Número de publicación | US7050491 B2 |

Tipo de publicación | Concesión |

Número de solicitud | US 10/271,386 |

Fecha de publicación | 23 May 2006 |

Fecha de presentación | 15 Oct 2002 |

Fecha de prioridad | 15 Oct 2001 |

Tarifa | Caducada |

También publicado como | US20030072363 |

Número de publicación | 10271386, 271386, US 7050491 B2, US 7050491B2, US-B2-7050491, US7050491 B2, US7050491B2 |

Inventores | James Douglas McDonald, Allen Le Roy Limberg |

Cesionario original | Mcdonald James Douglas, Allen Le Roy Limberg |

Exportar cita | BiBTeX, EndNote, RefMan |

Citas de patentes (7), Citada por (18), Clasificaciones (21), Eventos legales (4) | |

Enlaces externos: USPTO, Cesión de USPTO, Espacenet | |

US 7050491 B2

Resumen

An adaptive equalizer for amplitude-modulation (AM) signal received by over-the-air radio transmission comprises feed-forward finite-impulse-response (FIR) filtering and further comprises infinite-impulse-response (IIR) filtering for feeding back decisions. The echo content of baseband signal demodulated from the AM signal is measured for initially determining the reception channel impulse response (CIR) in the time-domain. Thereafter, the CIR is updated by decision feedback. Periodically, the CIR is normalized with respect to the strength of a cursor component thereof. The strengths of longer-delay post-echoes in the normalized CIR are used to determine the weighting coefficients of the IIR filtering, which suppresses the longer-delay post-echoes. The pre-echo and short-delay post-echo portion of the normalized CIR is convolved with the normalized CIR to generate a synthetic normalized CIR. The strengths of pre-echoes and short-delay post-echoes in the synthetic normalized CIR are used to determine the weighting coefficients of the feed-forward FIR filtering, which suppresses the pre-echoes and short-delay post-echoes.

Reclamaciones(20)

1. A method for adapting adjustable weighting coefficients of a feed-forward finite-impulse-response (FIR) digital filter that provides for performing at least a part of the channel-equalization and echo-suppression filtering of successive samples of a demodulation signal resulting from demodulation of an amplitude-modulation signal descriptive of digital symbols occurring at a baud rate, which demodulation signal is subject to being accompanied by unwanted echoes because of said amplitude-modulation signal being received via a channel that at times has more than one length of propagation path therethrough, which unwanted echoes are to be suppressed in successive baud-rate samples of a baseband response of said channel-equalization and echo-suppression filtering to said successive samples of said demodulation signal, said method comprising steps of:

initially measuring the echo content of samples of said demodulation signal to determine an initial channel impulse response (CIR) that characterizes the reception channel in the time-domain;

measuring reception error in said baseband response of said channel-equalization and echo-suppression filtering to said successive samples of said demodulation signal;

continuously updating the CIR responsive to said reception error;

periodically, normalizing the CIR with respect to the strength of a cursor component thereof to generate a normalized CIR;

convolving the normalized CIR with at least a portion thereof to generate a synthetic normalized CIR; and

determining the weighting coefficients of said feed-forward FIR digital filter from corresponding terms of said synthetic normalized CIR.

2. The method set forth in claim 1 , wherein said step of convolving the normalized CIR with at least a portion thereof to generate a synthetic normalized CIR is performed so as to convolve the normalized CIR with itself, so said feed-forward FIR digital filter provides for performing all of said channel-equalization and echo-suppression filtering.

3. The method set forth in claim 1 , wherein said step of convolving the normalized CIR with at least a portion thereof to generate a synthetic normalized CIR is performed so as to convolve the normalized CIR with just the portion thereof consisting of pre-echo terms and short-delay post-echo terms, so said feed-forward FIR digital filter provides for performing only a part of said channel-equalization and echo-suppression filtering, said method further adapting adjustable weighting coefficients of an infinite-impulse-response (IIR) digital filter that provides for performing a further part of said channel-equalization and echo-suppression filtering, said further adapting being done by the further step of:

determining the weighting coefficients of said IIR digital filter from long-delay post-echo terms of said normalized CIR.

4. Channel-equalization and echo-suppression filtering apparatus connected for processing successive samples of a demodulation signal resulting from demodulation of an amplitude-modulation signal descriptive of digital symbols occurring at a baud rate, which demodulation signal is subject to being accompanied by unwanted echoes because of said amplitude-modulation signal being received via a channel that at times has more than one length of propagation path therethrough, which unwanted echoes are to be suppressed in successive baud-rate samples of a baseband response of said channel-equalization and echo-suppression filtering apparatus to said successive samples of said demodulation signal, said channel-equalization and echo-suppression filtering apparatus comprising:

adaptive digital transversal filtering apparatus with adjustable weighting coefficients, said adaptive digital transversal filtering apparatus connected for receiving as an input signal thereof said successive samples of said demodulation signal and for supplying an output signal from which is derived said response of said channel-equalization and echo-suppression filtering apparatus to said successive samples of said demodulation signal;

apparatus for determining an initial channel impulse response (CIR) that characterizes the reception channel in the time-domain;

apparatus for generating estimates of the digital modulation of the amplitude-modulation signal as originally transmitted;

a reception error detector connected for generating a reception error signal responsive to the amount by which the baseband response of said channel-equalization and echo-suppression filtering apparatus differs from estimates of the digital modulation of the amplitude-modulation signal at the transmitter thereof;

a filter coefficients computer connected for receiving said CIR and said reception error signal, operable for continually updating the CIR responsive to said reception error signal, operable for periodically normalizing the CIR with respect to a principal component thereof to generate a periodically-updated normalized CIR, operable for computing an update of the adjustable weighting coefficients of said adaptive digital transversal filtering apparatus from each periodically-updated normalized CIR.

5. The channel-equalization and echo-suppression filtering apparatus of claim 4 , wherein said adaptive digital transversal filtering apparatus comprises:

apparatus for supplying said demodulation signal as a complex demodulation signal with orthogonal first and second components each sampled at twice baud rate, each of said orthogonal first and second components of said complex demodulation signal being supplied as even and odd sets of alternate samples occurring at baud rate an in temporal alignment with each other;

first, second, third and fourth adaptive filters of similar construction with respective sets of adjustable weighting coefficients that said filter coefficients computer adjusts in parallel, said first adaptive filter connected to respond to said odd set of alternate samples of said first component of said complex demodulation signal to generate a first adaptive filter response, said second adaptive filter connected to respond to said odd set of alternate samples of said second component of said complex demodulation signal to generate a second adaptive filter response, said third adaptive filter connected to respond to said even set of alternate samples of said first component of said complex demodulation signal to generate a third adaptive filter response, and said fourth adaptive filter connected to respond to said even set of alternate samples of said second component of said complex demodulation signal to generate a fourth adaptive filter response;

circuitry connected for supplying odd samples of I signal in a de-rotated response to said first adaptive filter response and said second adaptive filter response;

circuitry connected for supplying even samples of I signal in a de-rotated response to said third adaptive filter response and said fourth adaptive filter response; and

a first symbol synchronizer connected for receiving said odd samples of I signal from said circuitry for supplying them, connected for receiving said even samples of I signal from said circuitry for supplying them, and connected for supplying optimally sampled baud-rate samples of I signal as said baseband response of said channel-equalization and echo-suppression filtering apparatus.

6. The channel-equalization and echo-suppression filtering apparatus of claim 5 , wherein said adaptive digital transversal filtering apparatus includes:

first I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying odd samples of I signal,

circuitry connected for supplying odd samples of Q signal in a de-rotated response to said first adaptive filter response and said second adaptive filter response;

second I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying even samples of I signal,

circuitry connected for supplying even samples of Q signal in a de-rotated response to said third adaptive filter response and said fourth adaptive filter response; and

a second symbol synchronizer connected for receiving said odd samples of Q signal from said circuitry for supplying them, connected for receiving said even samples of Q signal from said circuitry for supplying them, and connected for supplying optimally sampled baud-rate samples of Q signal as a further response of said channel-equalization and echo-suppression filtering, said further response being supplied for controlling the demodulation of said amplitude-modulation signal.

7. The channel-equalization and echo-suppression filtering apparatus of claim 5 ; wherein said first, second, third and fourth adaptive filters of similar construction each comprise

a respective component FIR digital filter and a respective component IIR digital filter; wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component IIR digital filters from post-echo terms of said normalized CIR; wherein said filter coefficients computer is operable for convolving said normalized CIR with a portion of said normalized CIR consisting of its pre-echo terms to generate a synthetic normalized CIR; and wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component FIR digital filters from terms of said synthetic normalized CIR.

8. The channel-equalization and echo-suppression filtering apparatus of claim 5 ; wherein said first, second, third and fourth adaptive filters of similar construction each comprise

a respective component FIR digital filter and a respective component IIR digital filter; wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component IIR digital filters from longer-delayed post-echo terms of said normalized CIR; wherein said filter coefficients computer is operable for convolving said normalized CIR with a portion of said normalized CIR consisting of its pre-echo and shorter-delayed post-echo terms to generate a synthetic normalized CIR; and wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component FIR digital filters from at least the pre-echo and shorter-delayed post-echo terms of said synthetic normalized CIR.

9. The channel-equalization and echo-suppression filtering apparatus of claim 5 , wherein said first symbol synchronizer comprises:

I-signal interpolation circuitry for combining concurrent odd and even samples of said I signal as received by said first symbol synchronizer, said combining being done in a plurality of various ν and (1−ν) proportions to generate a plurality of respective I-signal interpolation results, where ν is a variable having a number of values from zero to one;

circuitry for detecting the departures of each of said plurality of respective I-signal interpolation results from estimates of transmitted symbols and determining respective averages of the absolute values of those departures of each of said plurality of respective I-signal interpolation results over a prescribed time interval; and

I-signal selection circuitry for selecting one of the respective I-signal interpolation results that has the lowest of said respective averages to be said baseband response of said channel-equalization and echo-suppression filtering apparatus.

10. The channel-equalization and echo-suppression filtering apparatus of claim 9 , wherein said adaptive digital transversal filtering apparatus includes:

first I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying odd samples of I signal,

circuitry connected for supplying odd samples of Q signal in a de-rotated response to said first adaptive filter response and said second adaptive filter response;

second I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying even samples of I signal,

circuitry connected for supplying even samples of Q signal in a de-rotated response to said third adaptive filter response and said fourth adaptive filter response;

Q-signal interpolation circuitry for combining concurrent ones of said odd and even samples of said Q signal, said combining being done in said plurality of various ν and (1−ν) proportions to generate a plurality of respective Q-signal interpolation results; and

Q-signal selection circuitry for selecting, as a signal for controlling the demodulation of said amplitude-modulation signal, the respective Q-signal interpolation result that combines said odd and even samples of said Q signal in the same ν and (1−ν) proportion as the I-signal interpolation result that said I-signal selection circuitry selects as said response of said channel-equalization and echo-suppression filtering apparatus.

11. The channel-equalization and echo-suppression filtering apparatus of claim 4 , wherein said adaptive digital transversal filtering apparatus comprises:

apparatus for supplying said demodulation signal as a complex demodulation signal with orthogonal first and second components each sampled at twice baud rate;

first and second adaptive filters of similar construction with respective sets of adjustable weighting coefficients that said filter coefficients computer adjusts in parallel, each said set of adjustable weighting coefficients being associated with kernel taps at two half-symbol-epochs spacing, the intervening kernel taps of said first and second adaptive filters being zero-valued, said first adaptive filter connected to respond to samples of said first component of said complex demodulation signal to generate a first adaptive filter response, said second adaptive filter connected to respond to samples of said second component of said complex demodulation signal to generate a second adaptive filter response;

first sample-separation circuitry for separating said first adaptive filter response into odd and even sets of alternate samples occurring at baud rate and being in temporal alignment with each other;

second sample-separation circuitry for separating said second adaptive filter response into odd and even sets of alternate samples occurring at baud rate and being in temporal alignment with each other;

circuitry connected for supplying odd samples of I signal in a de-rotated response to concurrent samples in the odd set of alternate samples of said first adaptive filter response and in the odd set of alternate samples of said second adaptive filter response;

circuitry connected for supplying even samples of I signal in a de-rotated response to concurrent samples in the even set of alternate samples of said first adaptive filter response and in the even set of alternate samples of said second adaptive filter response; and

a first symbol synchronizer connected for receiving said odd samples of I signal from said circuitry for supplying them, connected for receiving said even samples of I signal from said circuitry for supplying them, and connected for supplying optimally sampled baud-rate samples of I signal as said baseband response of said channel-equalization and echo-suppression filtering apparatus.

12. The channel-equalization and echo-suppression filtering apparatus of claim 11 , wherein said adaptive digital transversal filtering apparatus includes:

first I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying odd samples of I signal,

circuitry connected for supplying odd samples of Q signal in a de-rotated response to concurrent odd samples of said first adaptive filter response and said second adaptive filter response;

second I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying even samples of I signal,

circuitry connected for supplying even samples of Q signal in a de-rotated response to concurrent even samples of said first adaptive filter response and said second adaptive filter response;

a second symbol synchronizer connected for receiving said odd samples of Q signal from said circuitry for supplying them, connected for receiving said even samples of Q signal from said circuitry for supplying them, and connected for supplying optimally sampled baud-rate samples of Q signal as a further response of said channel-equalization and echo-suppression filtering apparatus, said further response being supplied for controlling the demodulation of said amplitude-modulation signal.

13. The channel-equalization and echo-suppression filtering apparatus of claim 11 ; wherein said first and second adaptive filters of similar construction each comprise

a respective component FIR digital filter and

a respective component IIR digital filter; wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component IIR digital filters from post-echo terms of said normalized CIR;

wherein said filter coefficients computer is operable for convolving said normalized CIR with a portion of said normalized CIR consisting of its pre-echo terms to generate a synthetic normalized CIR; and wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component FIR digital filters from terms of said synthetic normalized CIR.

14. The channel-equalization and echo-suppression filtering apparatus of claim 11 ; wherein said first and second adaptive filters of similar construction each comprise

a respective component FIR digital filter and

a respective component IIR digital filter; wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component IIR digital filters from longer-delayed post-echo terms of said normalized CIR; wherein said filter coefficients computer is operable for convolving said normalized CIR with a portion of said normalized CIR consisting of its pre-echo and shorter-delayed post-echo terms to generate a synthetic normalized CIR; and wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component FIR digital filters from at least the pre-echo and shorter-delayed post-echo terms of said synthetic normalized CIR.

15. The channel-equalization and echo-suppression filtering apparatus of claim 11 , wherein said first symbol synchronizer comprises:

I-signal interpolation circuitry for combining concurrent odd and even samples of said I signal as received by said first symbol synchronizer, said combining being done in a plurality of various ν and (1−ν) proportions to generate a plurality of respective I-signal interpolation results, where ν is a variable having a number of values from zero to one;

circuitry for detecting the departures of each of said plurality of respective I-signal interpolation results from estimates of transmitted symbols and determining respective averages of the absolute values of those departures of each of said plurality of respective I-signal interpolation results over a prescribed time interval; and

I-signal selection circuitry for selecting one of the respective I-signal interpolation results that has the lowest of said respective averages to be said baseband response of said channel-equalization and echo-suppression filtering apparatus.

16. The channel-equalization and echo-suppression filtering apparatus of claim 15 , wherein said adaptive digital transversal filtering apparatus includes:
first I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying odd samples of I signal,
second I-and-Q-extraction circuitry comprising, in addition to said circuitry connected for supplying even samples of I signal,

circuitry connected for supplying odd samples of Q signal in a de-rotated response to concurrent odd samples of said first adaptive filter response and said second adaptive filter response;

circuitry connected for supplying even samples of Q signal in a de-rotated response to concurrent even samples of said first adaptive filter response and said second adaptive filter response;

circuitry connected for supplying even samples of Q signal in a de-rotated response to said third adaptive filter response and said fourth adaptive filter response;

Q-signal interpolation circuitry for combining concurrent ones of said odd and even samples of said Q signal, said combining being done in said plurality of various ν and (1−ν) proportions to generate a plurality of respective Q-signal interpolation results; and

Q-signal selection circuitry for selecting, as a signal for controlling the demodulation of said amplitude-modulation signal, the respective Q-signal interpolation result that combines said odd and even samples of said Q signal in the same ν and (1−ν) proportion as the I-signal interpolation result that said I-signal selection circuitry selects as said response of said channel-equalization and echo-suppression filtering apparatus.

17. The channel-equalization and echo-suppression filtering apparatus of claim 4 ; wherein said adaptive digital transversal filtering apparatus is connected for receiving as an input signal thereof said demodulation signal as a real-only signal supplied at twice baud rate; wherein the adjustable weighting coefficients of said adaptive digital transversal filtering apparatus are associated with kernel taps at two half-symbol-epochs spacing, the intervening kernel taps of said adaptive digital transversal filtering apparatus being zero-valued; and wherein there is further included:

sample-separation circuitry for separating said output signal of said adaptive digital transversal filtering into odd and even sets of alternate samples occurring at baud rate and being in temporal alignment with each other

into a set of odd alternate samples thereof and a set of even alternate samples thereof, the samples of each of said sets of alternate samples occurring at baud rate and being in temporal alignment with each other; and

a symbol synchronizer connected for receiving said odd set of alternate samples of said output signal of said adaptive digital transversal filtering from said sample-separation circuitry that occur at baud rate, connected for receiving said even set of alternate samples of said output signal of said adaptive digital transversal filtering apparatus from said sample-separation circuitry, and connected for supplying optimally sampled baud-rate samples of said baseband response of said channel-equalization and echo-suppression filtering apparatus.

18. The channel-equalization and echo-suppression filtering apparatus of claim 17 ; wherein said adaptive digital transversal filtering apparatus comprises

a component FIR digital filter and a component IIR digital filter; wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component IIR digital filter from post-echo terms of said normalized CIR; wherein said filter coefficients computer is operable for convolving said normalized CIR with a portion of said normalized CIR consisting of its pre-echo terms to generate a synthetic normalized CIR; and wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component FIR digital filter from terms of said synthetic normalized CIR.

19. The channel-equalization and echo-suppression filtering apparatus of claim 17 ; wherein said adaptive digital transversal filtering apparatus comprises

a component FIR digital filter and a component IIR digital filter; wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component IIR digital filter from longer-delayed post-echo terms of said normalized CIR; wherein said filter coefficients computer is operable for convolving said normalized CIR with a portion of said normalized CIR consisting of its pre-echo and shorter-delayed post-echo terms to generate a synthetic normalized CIR; and wherein said filter coefficients computer is operable for computing the updates of the adjustable weighting coefficients of said component FIR digital filter from at least the pre-echo and shorter-delayed post-echo terms of said synthetic normalized CIR.

20. The channel-equalization and echo-suppression filtering apparatus of claim 17 , wherein said symbol synchronizer comprises:

interpolation circuitry for combining concurrent samples from said odd and even sets of alternate samples of said output signal of said adaptive digital transversal filtering, said combining being done in a plurality of various ν and (1−ν) proportions to generate a plurality of respective interpolation results, where ν is a variable having a number of values from zero to one;

circuitry for detecting the departures of each of said plurality of respective interpolation results from estimates of transmitted symbols and determining respective averages of the absolute values of those departures of each of said plurality of respective interpolation results over a prescribed time interval; and

selection circuitry for selecting one of the respective interpolation results that has the lowest of said respective averages to be said baseband response of said channel-equalization and echo-suppression filtering apparatus.

Descripción

This application is filed under 35 U.S.C. 111(a) claiming, pursuant to 35 U.S.C. 119(e)(1), benefit of the filing date of provisional U.S. patent application Ser. No. 60/329,424 filed Oct. 15, 2001, pursuant to 35 U.S.C. 111(b).

The invention relates to adaptive channel-equalization and echo-suppression filtering of digital modulating signal recovered from an amplitude-modulation signal, such as a broadcast digital television signal, that is subject to multipath distortion.

The broadcasting of digital television (DTV) signal in the United States has been done in accordance with a Digital Television Standard published in 1995 by the Advanced Television Systems Committee (ATSC) as Document A/53. An eight-level digital modulating signal controls the generation of a vestigial-sideband (VSB) signal with a suppressed very-high-frequency (VHF) or ultra-high-frequency (UHF) natural carrier, which VSB signal is transmitted together with a fixed-amplitude pilot carrier corresponding in frequency and phase with the suppressed natural carrier. The channel through which this VSB signal is transmitted with accompanying pilot carrier from the radio-frequency transmitter through the ether to the receiver is apt to include a number of component paths. Presuming there is no intervening barrier to transmission, the shortest of these paths is a direct line-of-sight path. Usually the channel will comprise a number of longer paths that result from the reflection of transmitted signal from objects outside the line-of-sight path. Multipath reception is a condition occurring when the channel includes a number of different paths, more than one of which contains sufficient energy to affect the recovery of digital modulating signal at the receiver.

The component of the broadcast DTV signal to which a DTV receiver synchronizes its operations is called the principal signal, and the principal signal is usually the strongest component of the broadcast DTV signal. The direct line-of-sight path is usually the strongest component of the broadcast DTV signal, if the direct line-of-sight path is not blocked by any intervening barrier to transmission. Therefore, the multipath signal components of the broadcast TV signal received over other paths are usually delayed with respect to the principal signal and appear as lagging multipath signals. It is possible however, that the direct or shortest path signal is not the signal to which the receiver synchronizes. When the receiver synchronizes its operations to a (longer path) signal that is delayed respective to the direct signal, there will be a leading multipath component caused by the direct signal. There may also be other leading signals caused by other reflected signals of lesser delay than the signal to which the receiver synchronizes. In the DTV art the multipath components of received signal are customarily referred to as “echoes”, because of their similarity to echoes in transmission lines that are terminated other than with their characteristic impedance. The leading multipath components are referred to as “pre-echoes”, and the lagging multipath components are referred to as “post-echoes”. The echoes vary in number, amplitude and delay time from location to location and from channel to channel at a given location. Post-echoes with significant energy have been reported as being delayed from the reference signal by as many as sixty microseconds. Pre-echoes with significant energy have been reported leading the reference signal by as many as thirty microseconds. Because of these variations in echo conditions adaptive filtering is used for suppressing multipath components other than the principal signal. Such an adaptive filter is commonly referred to as an “adaptive channel-equalization filter” or an “adaptive echo-suppression filter” or, more simply as an “adaptive equalizer” or “adaptive echo suppressor”. In this specification adaptive filtering used for channel equalization and echo suppression will be referred to as “adaptive equalizer” or just “equalizer”. The adaptive filtering is customarily digital filtering. The adaptive filtering can be performed on the IF DTV signal, if the IF DTV signal is digitized. However, in most designs the adaptive filtering is performed on the digital baseband DTV signal.

The approach generally followed in DTV receiver design is down-conversion of the radio-frequency (RF) DTV signal to an intermediate-frequency (IF) DTV signal and synchronous detection of the IF DTV signal to obtain a baseband DTV signal for application to the adaptive equalizer. In some designs the synchronous detection of the IF DTV signal is done in the analog regime, with the resulting analog baseband DTV signal being digitized for application to the adaptive equalizer. In other designs the IF DTV signal is digitized and synchronous detection of the digitized IF DTV signal is done in the digital regime, to generate the digital baseband DTV signal applied to the adaptive equalizer.

In some prior-art designs the adaptive equalizer is operative on a real-only baseband DTV signal. This signal is supplied from an in-phase synchronous detector that synchronously detects the IF DTV signal in accordance with a carrier that is synchronized with the suppressed carrier of the IF DTV signal and with the corresponding pilot carrier as converted to intermediate frequency. In many of these designs the real-only baseband DTV signal is sampled at twice Nyquist rate and the adaptive equalizer is of fractional type. In others of these designs, the adaptive equalizer is of synchronous type, with the real-only baseband DTV signal being subjected to a procedure known as phase-tracking before being sampled at Nyquist rate to supply input signal to the adaptive equalizer.

Other prior-art designs employ an adaptive equalizer that is complex in nature, being operative not only on a real component of baseband DTV signal supplied from an in-phase synchronous detector, but also on an imaginary component of baseband DTV signal supplied from an in-phase synchronous detector. J. D. McDonald and A. L. R. Limberg describe an alternative type of adaptive equalizer that is complex in nature in U.S. Pat. No. 6,975,689 issued 13 Dec. 2005 from application Ser. No. 09/823,500 filed 30 Mar. 2001. The patent and application are both titled “DIGITAL MODULATION SIGNAL RECEIVER WITH ADAPTIVE CHANNEL EQUALIZATION EMPLOYING DISCRETE FOURIER TRANSFORMS”. In this alternative type of adaptive equalizer synchronous detection is performed nominally at −45° phasing and at +45° phasing respective to the suppressed carrier of the IF DTV signal. The adaptive equalizer has respective portions for equalizing the response of the synchronous detector detecting at a nominally −45° carrier phase and for equalizing the response of the synchronous detector detecting at a nominally +45° carrier phase. The responses of these two portions of the adaptive equalizer are additively combined to recover a real baseband DTV signal and are differentially combined to recover an imaginary baseband DTV signal. The imaginary baseband DTV signal is lowpass filtered to develop an automatic phase control (APC) signal for the carrier generator generating −45°-phase and +45°-phase carrier signals used by the synchronous detectors. The result of performing synchronous detection at −45° phasing is the negative of the result of performing synchronous detection at +135° phasing as described herein.

U.S. Pat. No. 5,065,242 titled “DEGHOSTING APPARATUS USING PSEUDORANDOM SEQUENCES” issued 23 Aug. 1994 to Charles Dietrich and Arthur Greenberg. Pat. No. 5,065,242 describes the computation of the weighting coefficients of adaptive filters used for deghosting analog DTV signals, which computation is based on measurements of the channel impulse response (CIR) made during the reception of triple-PN127 sequences in a vertical-retrace-interval scan line. U.S. Pat. No. 6,975,689 describes adaptive equalizers in which weighting coefficients of the equalization filters are calculated from discrete Fourier transforms (DFTs) of data. Earlier adaptive equalizers for DTV receivers used auto-regression techniques based on the detection of reception error for adjusting the weighting coefficients of the adaptive equalizer by reducing the gradient of departures of equalizer response from ideal transmission symbols. In a procedure that is novel, the DTV receivers described in this specification use the detection of reception error for updating an initial measurement of CIR to track current reception conditions. Periodically, the weighting coefficients of the equalization filters are re-calculated based on a strobe of the continually updated CIR.

The synchronization of DTV receiver operations to the principal signal has been a source of long-standing problems for receiver designers. These problems have to do with developing an echo-free baseband DTV signal that is optimally sampled at baud rate, so each successive symbol of the recovered digital modulating signal experiences as little intersymbol interference (ISI) from preceding or succeeding samples as possible when subjected to data slicing procedures. Constraining the baseband DTV signal to Nyquist bandwidth makes the elimination of ISI possible, providing correct symbol synchronization can be achieved. That is, data slicing requires that sampling of the echo-free baseband DTV signal at the baud rate of the symbols be done in the exact phasing that minimizes ISI.

U.S. Pat. No. 5,479,449 titled “DIGITAL VSB DETECTOR WITH BANDPASS PHASE TRACKER, AS FOR INCLUSION IN AN HDTV RECEIVER” issued 26 Dec. 1995 to C. B. Patel and A. L. R. Limberg. U.S. Pat. No. 5,479,449 describes a symbol synchronization procedure that adjusts digital oversampling of a baseband DTV signal to optimize phasing for decimation to Nyquist rate. The oversampling samples preceding peak oversampling samples are accumulated, the oversampling samples succeeding peak oversampling samples are accumulated, and the accumulation results are compared to determine whether the oversampling clock should be advanced or retarded in phase so peak samples recovered at Nyquist rate through decimation are optimally phased. S. U. H. Qureshi in his paper “Timing Recovery for Equalized Partial-Response Systems”, IEEE TRANSACTIONS ON COMMUNICATIONS, December 1976, pp. 1326–1330 earlier described a similar method used for symbol synchronization of QAM signals. The adjustment of sample phasing moves the data-slicer window together with phasing of the Nyquist sample in these previously employed methods.

The DTV receivers described in this specification use a different method of symbol synchronization, in which the data-slicer window recurs at a predetermined phasing of the Nyquist sampling clock. A twice-Nyquist-rate sampling clock is determined from the symbol clock rate, as recovered by bright-line spectral recovery techniques. The digital baseband DTV signal is sampled in accordance with the twice-Nyquist-rate sampling clock. The oversampled digital baseband DTV signal is decimated by alternate sample selection techniques to generate two digital baseband DTV signals, each sampled at Nyquist rate by a sampling clock that is in staggered phasing with respect to the sampling clock of the other. These two digital baseband DTV signals, each sampled at Nyquist rate, are re-timed in accordance with the same Nyquist-rate sampling clock as will be used for data slicing and are then synchronously equalized independently of each other. The equalizer results are additively combined in preparation for data slicing, which additive combining is done in proportions subject to adjustment. The adjustment is made so that the additively combined equalizer results resample the synchronously equalized digital baseband DTV signal to be in temporal alignment with the data-slicer window that recurs at predetermined phasing of the Nyquist sampling clock. In effect, in DTV receivers described in this specification, the phase of the digital baseband DTV signal is adjusted vis-à-vis recurring data-slicer windows of predetermined phasing. This is done rather than following the practice in prior-art DTV receivers in which practice the data-slicer window is adjusted in phase to attempt to quantize the digital baseband DTV signal optimally.

The invention in various of its aspects is embodied in channel-equalization and echo-suppression filtering connected for processing successive samples of a demodulation signal resulting from demodulation of an amplitude-modulation signal descriptive of digital symbols occurring at a baud rate. The demodulation signal is subject to being accompanied by unwanted echoes because of the amplitude-modulation signal being received via a channel that at times has more than one length of propagation path therethrough. These unwanted echoes are to be suppressed in successive baud-rate samples of a baseband response of the channel-equalization and echo-suppression filtering to successive samples of the demodulation signal. The channel-equalization and echo-suppression filtering includes adaptive digital transversal filtering with adjustable weighting coefficients. The adaptive digital transversal filtering is connected for receiving as an input signal thereof successive samples of the demodulation signal and for supplying as an output signal thereof the response of the channel-equalization and echo-suppression filtering to those successive samples of the demodulation signal. The channel-equalization and echo-suppression filtering includes apparatus for determining an initial channel impulse response (CIR) that characterizes the reception channel in the time-domain, preferable forms of which apparatus are novel in the art. The channel-equalization and echo-suppression filtering includes apparatus for generating estimates of the digital modulation of the amplitude-modulation signal at the transmitter thereof and a reception error detector connected for generating a reception error signal responsive to the amount by which the baseband response of the channel-equalization and echo-suppression filtering differs from estimates of the digital modulation of the amplitude-modulation signal as originally transmitted. A filter coefficients computer for computing updated weighting coefficients for the adaptive digital transversal filtering by novel procedures is connected for receiving the initial CIR and the reception error signal. The filter coefficients computer is operable for continually updating the CIR responsive to said reception error signal, for periodically normalizing the CIR with respect to a principal component thereof to generate a periodically-updated normalized CIR, and for computing updated weighting coefficients for the adaptive digital transversal filtering from each periodically-updated normalized CIR.

**23** is further modified to incorporate a Costas loop for carrier recovery.

The **10** connected for receiving radio-frequency (RF) DTV signals from a receiving antenna **11**. The tuner **10** selects one of the RF DTV signals from the receiving antenna **11** or other source for conversion to an intermediate-frequency (IF) DTV signal. The tuner **10** provides gain for the selected RF DTV signal and for the IF DTV signal. The tuner **10** includes automatic gain control (AGC) that regulates the amplitude of the amplified IF DTV signal that the tuner **10** supplies to an analog-to-digital converter **12** as input signal thereto, so that the dynamic range of the ADC **12** is fully exploited. The ADC **12** samples the IF DTV signal at twice Nyquist rate and digitizes the samples to provide a digitized IF DTV signal, preferably one with at least 10 bits of resolution over full range of its response.

The ADC **12** is connected to apply digitized IF DTV signal as a respective input signal to each of two synchronous detectors **13** and **14**, which perform respective synchronous detection procedures nominally at +45° phasing and at +135° phasing respective to the suppressed carrier of the IF DTV signal. In actual practice the two synchronous detectors **13** and **14** are usually constructed as parts of a shared structure including digital complex multiplier apparatus. The digitized IF DTV signal is supplied to a phase-splitter in this shared structure for conversion to a complex digital multiplicand for the digital complex multiplier apparatus. A digital controlled oscillator (DCO) **15** is connected for supplying the digital complex multiplier apparatus with a digital complex multiplier signal. This digital complex multiplier signal is composed of two digital carrier signals orthogonal in their respective phasings. Nominally, one of these two digital carrier signals is advanced 45° respective to the suppressed carrier of the digitized IF DTV signal, and the other of these two digital carrier signals is advanced 135° respective to tile suppressed carrier of the digitized IF DTV signal. The demodulation results for the synchronous detectors **13** and **14** are the real and imaginary components, respectively, of the complex product signal generated by the digital complex multiplier apparatus in their shared structure. The demodulation results for the synchronous detectors **13** and **14** are each composed of a set of samples of baseband DTV signals generated at twice Nyquist rate. Each of these sets of samples may be considered as consisting of interleaved subsets of Nyquist-rate samples, one subset occurring at nτ times and the other subset occurring at (n+0.5)τ times, n being all integer values and τ being the periodicity of Nyquist-rate samples. The samples of the baseband DTV signals from the synchronous detectors **13** and **14** occurring at (n+0.5)τ times are applied as input signals to adaptive equalizers **16** and **17**, respectively, in a 2:1 decimation procedure. The samples of the baseband DTV signals from the synchronous detectors **13** and **14** are delayed 0.5τ by re-clocking in shim delay circuitry **18** and **19**, respectively. The re-clocked samples originally occurring at nτ times in the baseband DTV signals from the synchronous detectors **13** and **14** are applied as input signals to adaptive equalizers **20** and **21**, respectively, in a 2:1 decimation procedure. The re-clocking in delay circuitry **18** and in delay circuitry **19** temporally aligns the subsets of samples occurring at nτ times in the synchronous detector **13** and synchronous detector **14** responses that are applied to the adaptive equalizers **20** and **21**, respectively, with the subsets of samples occurring at (n+0.5)τ times in the synchronous detector **13** and synchronous detector **14** responses that are applied to the adaptive equalizers **16** and **17**, respectively. The operations of the adaptive equalizers **16**, **17**, **20** and **21** will be explained in detail further on in this specification.

I-and-Q-extraction circuitry **22** combines the responses of the adaptive equalizers **20** and **21** to generate an equalized in-phase (I) real component and an equalized quadrature-phase (Q) imaginary component of a baseband DTV signal, as expressed in samples originally occurring at nτ times. In simplest form, the I-and-Q-extraction circuitry **22** is essentially just a digital adder and a digital subtractor. The digital adder additively combines the responses of the adaptive equalizers **20** and **21** to generate the equalized imaginary (Q) component of the baseband DTV signal, as expressed in samples originally occurring at nτ times. The digital subtractor differentially combines the responses of the adaptive equalizers **20** and **21** to generate the equalized real (I) component of the baseband DTV signal, as expressed in samples occurring at nτ times. In more sophisticated form, as described further on in this specification with reference to **22** is a de-rotator circuit.

I-and-Q-extraction circuitry **23** combines the responses of the adaptive equalizers **16** and **17** to generate an equalized in-phase (I) real component and an equalized quadrature-phase (Q) imaginary component of a baseband DTV signal, as expressed in samples originally occurring at (n+0.5)τ times. In simplest form, the I-and-Q-extraction circuitry **23** is essentially just a digital adder and a digital subtractor. The digital adder additively combines the responses of the adaptive equalizers **16** and **17** to generate the equalized imaginary (Q) component of the baseband DTV signal, as expressed in samples originally occurring at (n+0.5)τ times. The digital subtractor differentially combines the responses of the adaptive equalizers **16** and **17** to generate the equalized real (I) component of the baseband DTV signal, as expressed in samples occurring at (n+0.5)τ times. In more sophisticated form, as described further on in this specification with reference to **23** is a de-rotator circuit.

A symbol synchronizer **24** combines the equalized real (I) component of the baseband DTV signal, as expressed in samples originally occurring at nτ times, with the equalized real (I) component of the baseband DTV signal, as expressed in samples originally occurring at (n+0.5)τ times, to generate the equalized real (I) component of a baseband DTV signal applied to a subsequent trellis decoder **25**. As will be explained in detail further on in this specification, the symbol synchronizer **24** combines the equalized real (I) components of the baseband DTV signal supplied from the I-and-Q-extraction circuitry **22** and from the I-and-Q-extraction circuitry **23** to generate the samples of the equalized real (I) component of baseband DTV signal applied to the trellis decoder **25**, so those samples are timed to occur at τ intervals that minimize inter-symbol interference (ISI).

A symbol synchronizer **26** combines the equalized imaginary (Q) component of the baseband DTV signal, as expressed in samples occurring at nτ times, with the equalized imaginary (Q) component of the baseband DTV signal, as expressed in samples occurring at (n+0.5)τ times, to generate an equalized imaginary (Q) component of a baseband DTV signal with samples timed to occur at the same τ intervals as the samples from the symbol synchronizer **24**. **24** to the symbol synchronizer **26** for timing the samples of the equalized imaginary (Q) component of the baseband DTV signal generated by the symbol synchronizer **26**. These samples are supplied to an AFPC loop filter **27**, which responds with an automatic-phase-and-frequency-control (AFPC) signal for the digital controlled oscillator **15**. The AFPC loop formed by feeding back the AFPC signal to the DCO **15** is designed to control the phases of the orthogonal digital carrier signals to keep synchrodyning of the digitized IF signal by the synchronous detector **13** advanced 45° and to keep synchrodyning of the digitized IF signal by the synchronous detector **14** advanced 135°, as compared to the suppressed carrier of the digitized IF DTV signal.

More particularly, the AFPC loop filter **27** includes a digital lowpass filter that has a time constant of at least a few milliseconds. The time constant of the AFPC loop is made long enough that it does not affect the adaptive equalization processes in the equalizers **16**, **17**, **20** and **21** appreciably. This avoids the feedback loops associated with adaptation of the weighting coefficients of digital filters in the equalizers **16**, **17**, **20** and **21** being de-stabilized by the AFPC loop controlling the phases of demodulation. The long time constant of the AFPC loop filter **27** means that rapid change in carrier phase owing to dynamic multipath reception conditions will not be tracked by the DCO **15** oscillations. The AFPC loop for the DCO **15** functions primarily as an automatic frequency control (AFC) loop. The equalized I signals generated by the I-and-Q-extraction circuitry **22** and **23**, however, do track changes in the amplitudes and phasing of spectral components that arise during dynamic multipath reception conditions, even rapid changes. So, the equalized I signal the symbol synchronizer **24** subsequently supplies to the trellis decoder **25** tracks those changes. The speed with which the weighting coefficients of the adaptive equalizers **16**, **17**, **20** and **21** can be adapted is determined primarily by the speed of the algorithms used for such adaptation. By design the oscillations from the DCO **15** are essentially of constant frequency and phase insofar as the adaptation of the weighting coefficients of the adaptive equalizers **16**, **17**, **20** and **21** is concerned.

Developing AFPC signal for the DCO **15** from equalized imaginary (Q) component of tie baseband DTV signal, rather than from an imaginary (Q) component of the baseband DTV signal before equalization, maintains the AFPC loop controlling the phases of demodulation under control of the principal component of received signal, rather than allowing it to be affected by variation in the other multipath components of received signal. Furthermore, if the component of received signal chosen as principal signal is changed with regard to computation of the equalized real (I) component of the baseband DTV signal, the newly chosen principal signal can immediately be used in the computation of the equalized quadrature (Q) component of the baseband DTV signal, with lessened discontinuity in the operation of the AFPC loop controlling the phases of demodulation.

A trellis coder **28** generates estimates of the transmitted symbols based on information it receives from the trellis decoder **25**. These estimates of the transmitted symbols are used in the adaptation of the weighting coefficients of digital filters in the equalizers **16**, **17**, **20** and **21**. This aspect of DTV receiver operation will be described in detail further on in this specification.

Otherwise, the operation of the **25** can be substantially in accordance with prior-art practice, as described in “Guide to the Use of the ATSC Digital Television Standard” that the ATSC published in 1995 as Document A/54. The trellis decoder **25** performs a 12-phase symbol decoding procedure to recover data. A de-interleaver **29** is connected to receive the recovered data and to assemble them into bytes that are then convolutionally de-interleaved. Circuitry **30** for Reed-Solomon error detection and correction is connected to receive the de-interleaved bytes of data supplied from the de-interleaver **29**. A data de-randomizer **31** is connected to receive the corrected data from the circuitry **30** and to supply de-randomized corrected data as a recovered transport stream to a transport stream de-multiplexer **32**. The transport stream de-multiplexer **32** sorts packets from the transport stream to the rest **33** of the receiving system in accordance with packet identifier (PID) information in the data stream. The transport stream de-multiplexer **32** also discards data the circuitry **30** indicates could not be corrected.

**16**, **17**, **20** and **21** can be operated. The baseband DTV signal supplied from the synchronous detector **13** or **14**, as possibly delayed by shim delay circuitry **18** or **19**, provides input signal to a buffer memory **35**, which can temporarily store a few thousand symbol epochs of baseband DTV signal.

The adaptive equalizer configuration commonly used in prior-art adaptive equalizers cascades an infinite-impulse-response (IIR) digital filter for suppressing post-echoes after a finite-impulse-response (FIR) digital filter for suppressing pre-echoes. In a customary procedure known as “decision feedback”, the response of the IIR filter is quantized before being weighted and fed back to be recursively combined with the FIR filter response that the IIR filter receives as input signal. The quantizing of the IIR filter response suppresses the growth of noise in the IIR filtering process. The suppression of a post-echo using an IIR filter does not generate unsuppressed repeats of the original post-echo.

The suppression of a pre-echo using an FIR filter does generate unsuppressed repeats of the original echoes, however. This comes about because the FIR filter output signal is a weighted summation of differentially delayed responses to the FIR filter input signal. The pre-echo component of a more-delayed response to the FIR filter input signal is suppressed by combining that response with an attenuated less-delayed response to the FIR filter input signal, rather than with an echo-free signal. So, the echoes of the attenuated less-delayed response to the FIR filter input signal appear in the FIR filter response.

Delayed baseband DTV signal supplied from the buffer memory **35** is the adaptive equalizer input signal applied as multiplicand input signal to a bank **36** of digital multipliers, which can be constructed in read-only memory so as to exhibit minimal delay. Digital multipliers **36**-**0**, **36**-**1**, **36**-**2**, . . . **36**-(N−2), **36**-(N−1) and **36**-N in the bank **36** weight the adaptive equalizer input signal by respective weights W_{0}, W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }to generate products applied as successive summands to a chain-of-adders register **37** formed from a plurality of adders, N in number. The adders **37**-**0**, **37**-**1**, **37**-**2**, . . . **37**-(N−2), and **37**-(N−1) are successive clocked digital adders in the register **37**, which adders together with a final unclocked digital adder **37**-N form a tapped delay line for the variously weighted adaptive equalizer input signals. The bank **36** of multipliers and the chain-of-adders register **37** together function as an FIR digital filter for suppressing pre-echoes in the adder **37**-N sum output signal that is the output signal of the adaptive equalizer.

**38** connected for quantizing the equalizer output signal supplied from the symbol synchronizer **24** to generate estimates of the symbols actually transmitted, and each successive symbol so estimated is temporarily stored in a sample latch **39**. A single quantizer **38** can be shared by each of the adaptive equalizers **16**, **17**, **20** and **21**. If ones of the adaptive equalizers **16**, **17**, **20** and **21** share a single quantizer **38**, they can also share a sample latch **39** in common. The estimates of the symbols actually transmitted, as temporarily stored in the sample latch **39**, are applied as multiplicand input signal to a bank **40** of digital multipliers, N+1 in number, which can be constructed in read-only memory so as to exhibit minimal delay. Digital multipliers **40**-**1**, **40**-**2**, **40**-**3**, . . . **40**-(N−1), **40**-N and **40**-(N+1) in the bank **40** weight the estimates of the symbols actually transmitted by respective weights W_{+1}, W_{+2}, W_{+3}, . . . W_{+(N−1)}, W_{+N }and W_{+(N+1) }for application to taps in the chain-of-adders register **37**. These feedback connections complete an IIR digital filter configuration for the adder **37**-N sum output signal. The adder **37**-N sum output signal for the adaptive equalizer **16** or **17** is supplied to the I-and-Q-extraction circuitry **23**, which supplies the in-phase signal component of the signal to the symbol synchronizer **24**. The adder **37**-N sum output signal for the adaptive equalizer **20** or **21** is supplied to the I-and-Q-extraction circuitry **22**, which supplies the in-phase signal component of the signal to the symbol synchronizer **24**. Completing the feedback connections for the IIR digital filter configuration through the chain-of-adders register **37** is a departure from common practice. This departure from common practice is advantageous because the IIR filter cancels more than just the post-echoes of the component of the adder **37**-N sum output signal that results from the adaptive equalizer input signal weighted by W_{0 }and most delayed by the chain-of-adders register **37**. (The adaptive equalizer input signal weighted by W_{0 }corresponds with the principal received signal component in the **37**-N sum output signal that result from the adaptive equalizer input signals weighted by W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }and that are less delayed by the chain-of-adders register **37**. This significantly reduces the noise in the adder **37**-N sum output signal when the transmission/reception channel is Rayleigh or is a Ricean channel with pronounced post-echo content. Only the pre-echoes of the components of the adder **37**-N sum output signal that result from the adaptive equalizer input signal less delayed by the chain-of-adders register **37** survive to require additional FIR filtering for their suppression.

The IIR filtering done by the bank **40** of multipliers and the chain-of-adders register **37** is a non-linear filtering procedure owing to the samples in the adder **37**-N sum output signal being quantized by the quantizer **38**. These estimates of the symbols actually transmitted differ from the samples in the adder **37**-N sum output signal in that these estimates are essentially free of digitized noise components, such as Johnson noise and lower energy multipath responses. The sample latch **39** could be used to feed back the adder **37**-N sum output signal in a linear IIR filtering procedure, rather than feeding back quantized adder **37**-N sum output signal in a non-linear filtering procedure. However, the adaptive-equalizer output signal supplied as adder **37**-N sum output signal exhibits less noise using the non-linear filtering procedure.

The adaptive equalizers **16**, **17**, **20** and **21** share a weighting coefficients computer **41**, a training signal gate **42**, CIR extraction circuitry **43**, a reception-error detector **45**, a symbol multiplexer **46**, a read-only memory **47** for training signal, and digital multiplier circuitry **48**. The weighting coefficients computer **41** computes the weighting coefficients W_{0}, W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }that the digital multipliers **36**-**0**, **36**-**1**, **36**-**2**, . . . **36**-(N−2), **36**-(N−1) and **36**-N respectively use as their multiplier input signals. The weighting coefficients computer **41** also computes the weighting Coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(N−1)}, W_{+N }and W_{+(N+1) }that the digital multipliers **40**-**1**, **40**-**2**, **40**-**3**, . . . **40**-(N−1), **40**-N and **40**-(N+1) respectively use as their multiplier input signals. The initial values of these weighting coefficients are determined in reliance upon training signal information contained in the baseband DTV signal.

**42** connected for selecting to CIR extraction circuitry **43** the training signal portions of the baseband DTV signals that the synchronous detectors **13** and **14** supply. The CIR extraction circuitry **43** extracts a complex transmission/reception channel impulse response (CIR) in the time-domain from these training signal portions selected by the training signal gate **42**. Initial values of the weighting coefficients applied to the set of samples processed by a particular adaptive equalizer are determined from the component of this complex CIR extracted from the training signal portion of its baseband DTV input signal.

By way of example, in a DTV signal transmitted in accordance with the A/53 standard, the training signal gate **42** selects at least the PN511 sequence of the initial data segment of each data field to the CIR extraction circuitry **43**. The CIR is forwarded from the CIR extraction circuitry **43** to the weighting coefficients computer **41**, which temporarily stores the CIR in an echo measurement register. A component of the temporarily stored CIR provides the basis for initializing the weighting coefficients of the **10** of

After initialization or re-initialization is completed, the computer **41** computes incremental adjustments to be made to the weighting coefficients used by the banks **36** and **40** of digital multipliers. The memory **35** delays the baseband DTV signal supplied from the synchronous detector for application as adaptive equalizer input signal. The delay is somewhat longer than the time for the weighting coefficients of the **35**. This delay permits the same training signal used for initialization or re-initialization to be supplied as a string of known symbols to the **41**, facilitating more rapid adjustment of any errors in the adaptive-equalizer weighting coefficients remnant after initialization or re-initialization.

More particularly, the memory **35** can temporarily store a sequence of samples from the final data segment of a data field and from the initial data segment of the next data field for recycling through the adaptive equalizer. If multipath reception conditions are slowly changing, this two-segment sequence can, be recycled through the adaptive equalizer not just once, but 156 times in a data field. This practice speeds up Wiener adaptation on the known DFS sequence appreciably, compared with Wiener adaptation relying on the known DFS sequence being available once per data field.

The in-phase adaptive equalizer output signal from the symbol synchronizer **24** is supplied to the reception-error detector **45**. The reception-error detector **45** is a digital subtractor that subtracts estimates of the originally transmitted data symbols supplied by the symbol multiplexer **46** from the in-phase adaptive equalizer output signal supplied by the symbol synchronizer **24**. The resulting difference is a reception-error signal, which the digital multiplier circuitry **48** multiplies by factor −μ to generate an incremental reception-error signal supplied to the weighting coefficients computer **41**. The number μ is customarily a small binary fraction.

During times the adaptive equalizer output signal responds to training signal portions of the delayed baseband DTV signal, the symbol multiplexer **46** selects samples of ideal training signal from the ROM **47** to the reception-error detector **45**. Then, the reception-error detector **45** generates a reception-error signal by differentially comparing the adaptive equalizer output signal with the samples of the ideal training signal from the ROM **47**. The digital multiplier **48** multiplies the reception-error signal by a negative multiplier signal, −μ, and is connected for supplying the resulting product to the weighting coefficients computer **41** as an incremental reception-error signal. The value μ is a scaling coefficient much less than unity.

At all other times, the symbol multiplexer **46** selects the response of the quantizer **38** to the reception-error detector **45** to provide the estimates of the originally transmitted data symbols. At these times the reception-error detector **45** functions as a decision-feedback-error detector. Then, the reception-error detector **45** generates a reception-error signal by differentially comparing the adaptive equalizer output signal with those estimates of the originally transmitted data symbols. The digital multiplier **48** multiplies the reception-error signal by the negative multiplier signal, −μ, and is connected for supplying the product signal to the weighting coefficients computer **41** as an incremental reception-error signal when training signal is not currently available.

The computer **41** is of a type for processing the incremental reception-error signal supplied from the digital multiplier **48** in accordance with known auto-regression techniques to compute incremental adjustments to be made to the weighting coefficients used as multiplier input signals by the banks **36** and **40** of digital multipliers. In accordance with ordinary practice, the computer **41** ascribes a higher confidence factor to samples of the reception-error signal generated by differentially comparing the adaptive equalizer output signal with the samples of the ideal training signal than the confidence factor the computer **41** ascribes to samples of the reception-error signal generated by differentially comparing the adaptive equalizer output signal with the estimates of the originally transmitted data symbols. However, in a departure from prior art practice the incremental adjustments are not made to the weighting coefficients W_{0}, W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }that the digital multipliers **36**-**0**, **36**-**1**, **36**-**2**, . . . **36**-(N−2), **36**-(N−1) and **36**-N respectively use as their multiplier input signals nor to the weighting coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(N−1)}, W_{+N }and W_{+(N+1) }that the digital multipliers **40**-**1**, **40**-**2**, **40**-**3**, . . . **40**-(N−1), **40**-N and **40**-(N+1) respectively use as their multiplier input signals. Instead, incremental adjustments are made to update the measured CIR, so the update CIR tracks current reception conditions. Periodically, the continuously updated CIR is strobed; and the weighting coefficients W_{0}, W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, W_{−N }and weighting coefficients W_{+1}, W_{+2}, W_{−3}, . . . W_{+(N−1)}, W_{+N }W_{+(N+1) }are re-calculated based on the strobed CIR.

**41** in detail. A CIR register **410** in the filter coefficients computer **41** stores the cepstrum—i.e., the transmission/reception channel impulse response in the time domain. This channel impulse response or CIR temporarily stored in the CIR register **410** is continually updated. Every P symbol epochs the CIR stored in the CIR register **410** at that time is used as the basis for generating a normalized CIR, in which the principal multipath component, or “cursor” component has +1 value.

**410** as a parallel-in, parallel-out (PIPO) register, the parallel loading of which is done via a multiplexer **411**. The CIR register **410** is commanded to clear its previous contents and load new contents, in an overwrite procedure which is done for each symbol processed by the **411** is conditioned to reproduce the CIR supplied from the **43**, for parallel loading into the cleared CIR register **410**. Then, a bank **412** of read-only-memory multipliers multiplies each of the terms of the CIR temporarily stored in the CIR register **410** by the reciprocal of the amplitude of the principal “cursor” term to generate a normalized CIR that is loaded into another parallel-in, parallel-out (PIPO) register **413**.

Thereafter, each symbol processed by the **45** to generate estimated error in each successive symbol. Circuitry **48** multiplies the estimated reception error in each successive symbol by an attenuation factor −μ to generate a factor that the bank **412** of ROM multipliers uses to multiply each term of the CIR stored in the CIR register **410**. This generates a respective incremental correction that will be used for updating the CIR temporarily stored in the CIR register **410**. Each of the terms of the CIR temporarily stored in the CIR register **410** and its incremental correction is supplied to a respective one of a bank **414** of clocked digital adders. The multiplexer **411** is conditioned to reproduce the sum output terms clocked forward from the bank **414** of digital adders, for accumulating the previous contents of the CIR register **410** with respective incremental corrections to update those previous contents.

Every P symbol epochs, in a step carried out between clock signals for the bank **414** of clocked digital adders, the bank **412** of read-only-memory multipliers multiplies each of the terms of the CIR then temporarily stored in the CIR register **410** by the reciprocal of the amplitude of the principal “cursor” term. This generates a normalized CIR that is loaded into the parallel-in, parallel-out (PIPO) register **413** for updating its contents. P is a number at least a few hundred and preferably not appreciably larger than half the number of terms in the discrete Fourier transforms that the filter coefficients computer **41** employs in its computation procedures. P is conveniently made equal to 832, since this facilitates comparing the CIR stored in the CIR register **414** at the beginning of a data field with the CIR extracted by the CIR extraction circuitry **43**. This comparison can be used as a basis for deciding whether or not the weighting coefficients in the

With a P of 832, the speed of tracking dynamic multipath conditions is faster than is generally needed. Larger values of P reduce processing speeds and conserve power. The A/53 data field has 313 data segments of 832 symbols. Since 313 is prime, P cannot be made a multiple of 832. However, supposing 4096-term DFTs are employed, P can be made 4 times 313, or 1252, to facilitate the CIR stored in the CIR register **414** at the beginning of a data field being compared on regular interval with the CIR extracted by the CIR extraction circuitry **43**. With slight risk of wrap-around effects in the DFTs, P may instead be made 8 times 313, or 2504. The A/53 Digital Television Standard could be modified to specify a larger number of data segments per data field in order to accommodate a longer training signal. If the A/53 data field were extended to an even number of data segments, then, supposing 4096-term DFTs are employed, P would conveniently be made twice 832, or 1664, to facilitate CIR comparisons being made regularly each data field. If the A/53 data field were extended to 315 data segments to accommodate a longer training signal, then, supposing 4096-term DFTs are employed, P could be made thrice 832, or 2496, to facilitate CIR comparisons being made regularly each data field.

The cursor term of the CIR temporarily stored in the CIR register **413** is applied as a W_{0 }weighting coefficient to the multiplier **37**-**0** in **22** or **23** of **413** are multiplied by minus one in circuitry **415**. This generates the weighting coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(N−1)}, W_{+N }and W_{+(N+1) }that the digital multipliers **40**-**1**, **40**-**2**, **40**-**3**, . . . **40**-(N−1), **40**-N and **40**-(N+1) respectively use as their multiplier input signals. The IIR filtering performed by the chain-of-adders register **37** and the bank **40** of digital multipliers cancels all post-echoes, but pre-echoes remain to be suppressed by FIR filtering performed by the bank **36** of digital multipliers and the chain-of-adders register **37**.

The problem is that the signals being differentially combined to cancel pre-echoes themselves are subject to pre-echoes of succeeding signals being included within them, rather than being echo-free. So the pre-echo terms of each CIR temporarily stored in the CIR register **413** cannot simply multiplied by minus unity for generating weights W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }to be applied as multiplier input signals to the digital multipliers **36**-**1**, **36**-**2**, . . . **36**-(N−2), **36**-(N−1) and **36**-N, respectively, in the **36** of ROM multipliers. The errors in such procedure are more pronounced if there is a larger number of echoes of substantial energy in the received DTV signal. The effect of these pre-echoes of succeeding signals is removed in accordance with an aspect of the invention. In this aspect of the invention the CIR is used to generate the deconvolution result for a DTV signal transmitted with flat frequency response and received through the actual transmission/reception channel. Incidentally, this deconvolution result generates the terms of an impulse response descriptive of an FIR filter that accounts for the echo-suppression components including post-echoes of preceding signals and pre-echoes of succeeding signals. While an FIR filter of this type could be employed as equalizer, the

Accordingly, in **416** computes the discrete Fourier transform (DFT) of each normalized CIR temporarily stored in the CIR register **413** to determine the channel impulse response in the frequency domain. This DFT provides the divisor terms for a bank **417** of read-only-memory dividers that perform a term-by-corresponding-term division process in the frequency domain, which process corresponds to a de-convolution process in the time domain. Term-by-corresponding-term division of a flat frequency-domain response (with all unity terms) by the DFT of the normalized CIR would generate the DFT of the overall response required of the **416** computes that would be smaller than a prescribed value associated with acceptably small noise growth are augmented by the computer **416**, so as to replace the originally computed value with that prescribed value.

The CIR register **413** is connected for supplying only the pre-echo terms of each normalized CIR temporarily stored therein to a component computer **418**. The component computer **418** computes the DFT of the then-current pre-echo terms for application to the bank **417** of ROM dividers as respective dividend terms for the term-by-corresponding-term division process in the frequency domain that corresponds to a de-convolution process in the time domain. The quotients from the bank **417** of ROM dividers specify a quotient DFT. A component computer **419** is connected to receive this DFT from the bank **417** of ROM dividers. The component computer **419** computes the inverse discrete Fourier transform (I-DFT) of the quotient DFT to synthesize the synthetic normalized CIR that is the impulse response in the time-domain that is to be compensated against by the FIR filter comprising the bank **36** of multipliers and the chain-of-adders register **37**. The echo terms of this synthetic normalized CIR are multiplied by minus one in circuitry **41**A to generate weights W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }respectively applied as multiplier input signals to the digital multipliers **36**-**1**, **36**-**2**, . . . **36**-(N−2), **36**-(N−1) and **36**-N in the **36** of ROM multipliers.

**22** following the adaptive equalizers **20** and **21** in the **20** and **21** supply respective output R_{20 }and R_{21 }responses. The adaptive equalizers **20** and **21** have respective W_{0 }weighting coefficients W_{0-20 }and W_{0-21 }in their respective channel impulse responses just before those CIRs are normalized. The reciprocals of these weighting coefficients W_{0-20 }and W_{0-21 }are used in the normalization of those CIRs.

The combined R_{20 }and R_{21 }responses are expressible as a phasor R_{20}+jR_{21}. The W_{0-20 }and W_{0-21 }weighting coefficients are expressible as a phasor W_{0-20}+jW_{0-21}. De-rotation of the phasor R_{20}+jR_{21 }is accomplished by dividing it by the phasor W_{0-20}+jW_{0-21 }to obtain the I signal.

*I=*(*R* _{20} *+jR* _{21})/(*W* _{0-20} *+jW* _{0-21}).

Multiplying by (W_{0-20}−jW_{0-21})/(W_{0-20}−jW_{0-21}):

*I=*(*W* _{0-20} *R* _{20} *−W* _{0-21} *R* _{21})/(*W* _{0-20} ^{2} *+W* _{0-21} ^{2}).

A read-only memory **221** is connected to receive the weighting coefficient W_{0-20 }and the R_{20 }response of adaptive equalizer **20** as respective partial input addresses and to supply their product W_{0-20}R_{20 }as minuend input signal to a digital subtractor **222**. A read-only memory **223** is connected to receive the weighting coefficient W_{0-21 }and the R_{21 }response of adaptive equalizer **21** as respective partial input addresses and to supply their product W_{0-21}R_{21 }as subtrahend input signal to the digital subtractor **222**. A read-only memory **224** is connected to respond to weighting coefficients W_{0-20 }and W_{0-21 }as respective partial input addresses to supply the reciprocal of the sum of their squares as its response. A read-only memory **225** is connected to receive the difference output signal (W_{0-20}R_{20}−W_{0-21}R_{21}) from the subtractor **222** as a partial input address for multiplication by the 1/(W_{0-20} ^{2}+W_{0-21} ^{2}) response of the ROM **224** received as the rest of its input address. The ROM digital multiplier **225** is connected to supply its (W_{0-20}R_{20}−W_{0-21}R_{21})/(W_{0-20} ^{2}+W_{0-21} ^{2}) product output signal as the I signal output of the I-and-Q-extraction circuitry **22**.

A phasor lagging the phasor W_{0-20}+jW_{0-21 }by 90° is:

*−j*(*W* _{0-20} *+jW* _{0-21})=*W* _{0-21} *−jW* _{0-20}.

Dividing the phasor R_{20}+jR_{21 }by the phasor W_{0-21}−jW_{0-20 }generates the Q signal.

*Q*=(*R* _{20} *+jR* _{21})/(*W* _{0-21} *−jW* _{0-20}).

Multiplying by (W_{0-21}−jW_{0-20})/(W_{0-21}−jW_{0-20}):

*Q*=(*W* _{0-21} *R* _{20} *+W* _{0-20} *R* _{21})/(*W* _{0-20} ^{2} *+W* _{0-21} ^{2}).

A read-only memory **226** is connected to receive the weighting coefficient W_{0-20 }and the R_{21 }response of adaptive equalizer **21** as respective partial input addresses and to supply their product W_{0-20}R_{21 }as a first summand to a digital adder **227**. A read-only memory **228** is connected to receive the weighting coefficient W_{0-21 }and the R_{20 }response of adaptive equalizer **20** as respective partial input addresses and to supply their product W_{0-21}R_{20 }as a second summand to the digital adder **227**. A read-only memory **229** is connected to receive the sum output signal (W_{0-21}R_{20}+W_{0-20}R_{21}) from the adder **227** as a partial input address for multiplication by the 1/(W_{0-20} ^{2}+W_{0-21} ^{2}) response of the ROM **224** received as the rest of its input address. The ROM digital multiplier **229** is connected to supply its (W_{0-21}R_{20}+W_{0-20}R_{21})/(W_{0-20} ^{2}+W_{0-21} ^{2}) product output signal as the Q signal output of the I-and-Q-extraction circuitry **22**.

**23** following the adaptive equalizers **16** and **17** in the **16** and **17** supply respective output R_{16 }and R_{17 }responses. The adaptive equalizers **16** and **17** have respective weighting coefficients W_{0-16 }and W_{0-17 }in their respective channel impulse responses just before those CIRs are normalized in their respective filter coefficient computers **41**. The reciprocals of these weighting coefficients W_{0-16 }and W_{0-17 }are used in the normalization of those CIRs.

A read-only memory **231** is connected to receive the weighting coefficient W_{0-16 }and the R_{16 }response of adaptive equalizer **16** as respective partial input addresses and to supply their product W_{0-16}R_{16 }as minuend input signal to a digital subtractor **232**. A read-only memory **233** is connected to receive the weighting coefficient W_{0-17 }and the R_{17 }response of adaptive equalizer **17** as respective partial input addresses and to supply their product W_{0-17}R_{17 }as subtrahend input signal to the digital subtractor **232**. A read-only memory **234** is connected to respond to weighting coefficients W_{0-16 }and W_{0-17 }as respective partial input addresses to supply the reciprocal of the sum of their squares as its response. A read-only memory **235** is connected to receive the difference output signal (W_{0-16}R_{16}−W_{0-17}R_{17}) from the subtractor **232** as a partial input address for nultiplication by the 1/(W_{0-16} ^{2}+W_{0-17} ^{2}) response of the ROM **234** received as the rest of its input address. The ROM digital multiplier **235** is connected to supply its (W_{0-16}R_{16}−W_{0-17}R_{17})/(W_{0-16} ^{2}+W_{0-17} ^{2}) product output signal as the I signal output of the I-and-Q-extraction circuitry **23**.

A read-only memory **236** is connected to receive the weighting coefficient W_{0-16 }and the R_{17 }response of adaptive equalizer **17** as respective partial input addresses and to supply their product W_{0-16}R_{17 }as a first summand to a digital adder **237**. A read-only memory **238** is connected to receive the weighting coefficient W_{0-17 }and the R_{16 }response of adaptive equalizer **16** as respective partial input addresses and to supply their product W_{0-17}R_{16 }as a second summand to the digital adder **237**. A read-only memory **239** is connected to receive the sum output signal (W_{0-17}R_{16}+W_{0-16}R_{17}) from the adder **237** as a partial input address for multiplication by the 1/(W_{0-16} ^{2}+W_{0-17} ^{2}) response of the ROM **234** received as the rest of its input address. The ROM digital multiplier **239** is connected to supply its (W_{0-17}R_{16}+W_{0-16}R_{17})/(W_{0-16} ^{2}+W_{0-17} ^{2}) product output signal as the Q signal output of the I-and-Q-extraction circuitry **23**.

In practice, the ROM digital multiplier **229** of **239** of **15** of _{0-21}R_{20}+W_{0-20}R_{21}) from the adder **227** of _{0-17}R_{16}+W_{0-16}R_{17}) from the adder **237** of **26** of

In effect, the procedures carried out in the I-and-Q-extraction circuitry **22** and in the I-and-Q-extraction circuitry **23** are phase rotation procedures for optimizing the phase of demodulation for best signal-to-noise ratio (SNR) of baseband DTV signal. The level of random noise in the I output signals from the I-and-Q-extraction circuitry **22** and **23** that is attributable to digitized Johnson noise from the **10** does not change as the effective phase of demodulation is changed. Nor does quantization noise from the **12**. SNR changes because the I output signals from the I-and-Q-extraction circuitry **22** and **23** have their largest amplitudes when the effective phase of demodulation is optimized.

**24** in the **25**, the symbol synchronizer **24** combines an input signal I_{22 }composed of equalized I samples from the I-and-Q-extraction circuitry **22** with an input signal I_{23 }composed of equalized samples from the I-and-Q-extraction circuitry **23**. I_{22 }and I_{23 }are combined in various ν and (1−ν) proportions, where ν is a variable having a value of zero, one-eighth, one-quarter; three-eighths, one-half, five-eighths, three-quarters, seven-eighths or one. Selection circuitry **2400** selects the one of the combined signals for application to the trellis decoder **25** that will generate the smallest amounts of decision-feedback error. The following three paragraphs describe generation of the signals combining I_{22 }and I_{23 }in various ν and (1−ν) proportions for selection by the selection circuitry **2400**.

The equalized in-phase signal I_{22 }from the I-and-Q-extraction circuitry **22** is divided by two, four and eight by digital dividers **2401**, **2402** and **2403** to generate I_{22}/2, I_{22}/4 and I_{22}/8 signals. The equalized in-phase signal I_{23 }from the I-and-Q-extraction circuitry **23** is divided by two, four and eight by digital dividers **2404**, **2405** and **2406** to generate I_{23}/2, I_{23}/4 and I_{23}/8 signals. Binary point shift division methods are the best way to construct the digital dividers **2401**, **2402**, **2403**, **2404**, **2405** and **2406**. A digital adder **2407** adds the I_{22}/4 and I_{22}/8 signals from the dividers **2402** and **2403** to generate 3I_{22}/8 signal; and a digital adder **2408** adds the I_{23}/4 and I_{23}/8 signals from the dividers **2405** and **2406** to generate 3I_{23}/8 signal. A digital adder **2409** adds the I_{22}/2 and I_{22}/8 signals from the dividers **2401** and **2403** to generate 5I_{22}/8 signal; and a digital adder **2410** adds the I_{23}/2 and I_{23}/8 signals from the dividers **2404** and **2406** to generate 5I_{23}/8 signal. A digital adder **2411** adds the I_{22}/2 and I_{22}/8 signals from the dividers **2401** and **2403** to generate 3I_{22}/4 signal; and a digital adder **2412** adds the I_{23}/2 and I_{23}/4 signals from the dividers **2404** and **2405** to generate 3I_{23}/4 signal. A digital subtractor **2413** subtracts from the I_{22 }signal the I_{22}/8 signal from the divider **2403** to generate 7I_{22}/8 signal; and a digital subtractor **2414** subtracts from the I_{23 }signal the I_{23}/8 signal from the divider **2406** to generate 7I_{23}/8 signal.

A clocked digital adder **2415** adds the 7I_{22}/8 difference output signal from the digital subtractor **2413** and the I_{23}/8 signal from the digital divider **2406** to generate a combined signal in which ν equals ⅞, for application to the selection circuitry **2400**. A clocked digital adder **2416** adds the 3I_{22}/4 sum output signal from the digital adder **2411** and the I_{23}/4 signal from the digital divider **2405** to generate a combined signal in which ν equals ¾, for application to the selection circuitry **2400**. A clocked digital adder **2417** adds the 5I_{22}/8 sum output signal from the digital adder **2409** and the 3I_{23}/8 sum output signal from the digital adder **2408** to generate a combined signal in which ν equals ⅝, for application to the selection circuitry **2400**. A clocked digital adder **2418** adds the I_{22}/2 and I_{23}/2 signals from the digital dividers **2401** and **2404** to generate a combined signal in which ν equals ½, for application to the selection circuitry **2400**. A clocked digital adder **2419** adds the 3I_{22}/8 sum output signal from the digital adder **2407** and the 5I_{23}/8 sum output signal from the digital adder **2410** to generate a combined signal in which ν equals ⅜, for application to the selection circuitry **2400**. A clocked digital adder **2420** adds the I_{22}/4 signal from the digital divider **2402** and the 3I_{23}/4 sum output signal from the digital adder **2412** to generate a combined signal in which ν equals ¼, for application to the selection circuitry **2400**. A clocked digital adder **2421** adds the I_{22}/8 signal from the digital divider **2403** and the 7I_{22}/8 difference output signal from the digital subtractor **2414** to generate a combined signal in which ν equals ⅛, for application to the selection circuitry **2400**.

Each successive sample of the equalized in-phase signal I_{22 }from the I-and-Q-extraction circuitry **22** is temporarily stored in a clocked sample latch **2422** to generate a signal in which ν equals unity, for application to the selection circuitry **2400**. The clocked sample latch **2422** temporally aligns this signal in which ν equals unity with the sum output signals of the clocked digital adders **2415**, **2416**, **2417**, **2418**, **2419**, **2420** and **2421**. Each successive sample of the equalized in-phase signal I_{23 }from the I-and-Q-extraction circuitry **23** is temporarily stored in a clocked sample latch **2422** to generate a signal in which ν equals zero, for application to the selection circuitry **2400**. The clocked sample latch **2423** temporally aligns this signal in which ν equals zero with the sum output signals of the clocked digital adders **2415**, **2416**, **2417**, **2418**, **2419**, **2420** and **2421**.

**26** in the **27**, the symbol synchronizer **26** combines an input signal Q_{22 }composed of equalized Q samples from the I-and-Q-extraction circuitry **22** with an input signal Q_{23 }composed of equalized samples from the I-and-Q-extraction circuitry **23**. Q_{22 }and Q_{23 }are combined in various ν and (1−ν) proportions, where ν is a variable having a value of zero, one-eighth, one-quarter, three-eighths, one-half, five-eighths, three-quarters, seven-eighths or one. Selection circuitry **2600** selects, as input signal to the AFPC loop filter **27**, the one of the combined signals that corresponds with the combined signal that the selection circuitry **2400** selects to the trellis decoder **25**. The following three paragraphs describe generation of the signals combining Q_{22 }and Q_{23 }in various ν and (1−ν) proportions for selection by the selection circuitry **2600**.

The equalized quadrature-phase signal Q_{22 }from the I-and-Q-extraction circuitry **22** is divided by two, four and eight by digital dividers **2601**, **2602** and **2603** to generate Q_{22}/2, Q_{22}/4 and Q_{22}/8 signals. The equalized quadrature-phase signal Q_{23 }from the I-and-Q-extraction circuitry **23** is divided by two, four and eight by digital dividers **2604**, **2605** and **2606** to generate Q_{23}/2, Q_{23}/4 and Q_{23}/8 signals. Binary point shift division methods are the best way to construct the digital dividers **2601**, **2602**, **2603**, **2604**, **2605** and **2606**. A digital adder **2607** adds the Q_{22}/4 and Q_{22}/8 signals from the dividers **2602** and **2603** to generate 3Q_{22}/8 signal; and a digital adder **2608** adds the Q_{23}/4 and Q_{23}/8 signals from the dividers **2605** and **2606** to generate 3Q_{23}/8 signal. A digital adder **2609** adds the Q_{22}/2 and Q_{22}/8 signals from the dividers **2601** and **2603** to generate 5Q_{22}/8 signal; and a digital adder **2610** adds the Q_{23}/2 and Q_{23}/8 signals from the dividers **2604** and **2606** to generate 5Q_{23}/8 signal. A digital adder **2611** adds the Q_{22}/2 and Q_{22}/8 signals from the dividers **2601** and **2603** to generate 3Q_{22}/4 signal; and a digital adder **2612** adds the Q_{23}/2 and Q_{23}/4 signals from the dividers **2604** and **2605** to generate 3Q_{23}/4 signal. A digital subtractor **2613** subtracts from the Q_{22 }signal the Q_{22}/8 signal from the divider **2603** to generate 7Q_{22}/8 signal; and a digital subtractor **2614** subtracts from the Q_{23 }signal the Q_{23}/8 signal from the divider **2606** to generate 7Q_{23}/8 signal.

A clocked digital adder **2615** adds the 7Q_{22}/8 difference output signal from the digital subtractor **2613** and the Q_{23}/8 signal from the digital divider **2606** to generate a combined signal in which ν equals ⅞, for application to the selection circuitry **2600**. A clocked digital adder **2616** adds the 3Q_{22}/4 sum output signal from the digital adder **2611** and the Q_{23}/4 signal from the digital divider **2605** to generate a combined signal in which ν equals ¾, for application to the selection circuitry **2600**. A clocked digital adder **2617** adds the 5Q_{22}/8 sum output signal from the digital adder **2609** and the 3Q_{23}/8 sum output signal from the digital adder **2608** to generate a combined signal in which ν equals ⅝, for application to the selection circuitry **2600**. A clocked digital adder **2618** adds the Q_{22}/2 and Q_{23}/2 signals from the digital dividers **2601** and **2604** to generate a combined signal in which ν equals ½, for application to the selection circuitry **2600**. A clocked digital adder **2619** adds the 3I_{22}/8 sum output signal from the digital adder **2607** and the 5Q_{23}/8 sum output signal from the digital adder **2610** to generate a combined signal in which ν equals ⅜, for application to the selection circuitry **2600**. A clocked digital adder **2620** adds the Q_{22}/4 signal from the digital divider **2602** and the 3Q_{23}/4 sum output signal from the digital adder **2612** to generate a combined signal in which ν equals ¼, for application to the selection circuitry **2600**. A clocked digital adder **2621** adds the Q_{22}/8 signal from the digital divider **2603** and the 7Q_{22}/8 difference output signal from the digital subtractor **2614** to generate a combined signal in which ν equals ⅛, for application to the selection circuitry **2600**.

Each successive sample of the equalized quadrature-phase signal Q_{22 }from the I-and-Q-extraction circuitry **22** is temporarily stored in a clocked sample latch **2622** to generate a signal in which ν equals unity, for application to the selection circuitry **2600**. The clocked sample latch **2622** temporally aligns this signal in which ν equals unity with the sum output signals of the clocked digital adders **2615**, **2616**, **2617**, **2618**, **2619**, **2620** and **2621**. Each successive sample of the equalized quadrature-phase signal Q_{23 }from the I-and-Q-extraction circuitry **23** is temporarily stored in a clocked sample latch **2622** to generate a signal in which ν equals zero, for application to the selection circuitry **2600**. The clocked sample latch **2623** temporally aligns this signal in which ν equals zero with the sum output signals of the clocked digital adders **2615**, **2616**, **2617**, **2618**, **2619**, **2620** and **2621**.

**2400**, which internal structure is used in the processing of one of he nine I input signals supplied to the selection circuitry **2400**. The successive samples of the I input signal are supplied to a preliminary quantizer **2424**, which estimates the transmitted symbols giving rise to the samples. A clocked digital subtractor **2425** determines the departures of the successive samples of the I input signal from the estimates of the transmitted symbols. Absolute-value circuitry **2426** determines the absolute values of these departures, which are supplied to accumulator circuitry **2427** that integrates over a fixed interval, which fixed interval is indicated in **2428** then ascertains whether the integration result obtained by the accumulator circuitry **2427** is the smallest of the integration results obtained for all nine of the I input signals supplied to the selection circuitry **2400**. If and only if the integration result obtained by the accumulator circuitry **2427** is determined to be the smallest of the integration results obtained for all nine of the I input signals supplied to the selection circuitry **2400**, a gate circuit **2429** is conditioned to forward the successive samples of the I input signal supplied to the preliminary quantizer **2424**. The gate circuit **2429** is connected for supplying the forwarded samples to the trellis decoder **25** as its input signal.

**2600**, which internal structure is used in the processing of one of the nine Q input signals supplied to the selection circuitry **2600**. Comparator circuitry **2428** in the selection circuitry **2400** is connected to control a gate circuit **2624** in the selection circuitry **2600**. If and only if the integration result obtained by the accumulator circuitry **2427** is determined to be the smallest of the integration results obtained for all nine of the I input signals supplied to the selection circuitry **2400**, the gate circuit **2624** is conditioned to forward the successive samples of the corresponding Q input signal samples to the AFPC loop filter **27** as its input signal.

As described earlier in this specification, the symbol synchronizer **24** re-samples twice-Nyquist-rate I samples to a number of streams of Nyquist-rate I samples in a number of parallel 2:1 decimation filtering steps performed in different phases. The symbol synchronizer **24** selects, as input to the trellis decoder **25**, the stream of Nyquist-rate I samples that is phased so as to minimize intersymbol interference (ISI) in data slicing performed in the trellis decoder **25**. This procedure avoids the interaction of a feedback loop for adjusting symbol synchronization with the feedback loops for adjusting weighting coefficients of the component adaptive synchronous equalizers **16**, **17**, **20** and **21**. There is no lag of a feedback loop for adjusting symbol synchronization to affect the feedback loops for adjusting weighting coefficients of the component adaptive synchronous equalizers **16**, **17**, **20** and **21**. Aspects of the invention are useful in equalizers that use types of symbol synchronization other than the preferred symbol synchronization described in this specification with reference to **7** and **8** of the drawing.

In preferred embodiments of the invention not all components of the IIR filtered decision feedback signal are generated from the response of a quantizer **38** supplying estimates of the symbols actually transmitted within a symbol epoch. Instead, the longer-delayed components of the IIR filtered decision feedback signal are generated from more reliable estimates of the symbols actually transmitted, which more reliable estimates are supplied from the trellis decoder **25**. These connections are too complicated to show explicitly in a drawing figure of reasonable size. The delay in each of the more reliable estimates becoming available must be compensated for in applying weighted decision-feedback terms to the chain-of-adders register **37**. The equalizer response is provided with delay that compensates for the delay in the most reliable estimates of the symbols actually transmitted becoming available, so the input signals for the reception-error detector are in proper temporal alignment with each other.

The description thusfar is intended to provide background for understanding the general design of the adaptive, complex, fractional equalizers that embody the invention. The preferred construction for the adaptive Asynchronous equalizers **16**, **17**, **20** and **21** differs somewhat from the pipeline structure shown in

**16**, **17**, **20** and **21** alternative to the structure shown in **41** is replaced by a different filter coefficients computer **49**, which includes component circuitry that **50** connected for temporarily storing 512 samples of the adaptive equalizer output signal supplied as sum output signal from the final unclocked digital adder **37**-N of the chain-of-adders register **37**. The **36** of digital multipliers used in the _{0}, W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }to generate products applied as successive summands to the chain-of-adders register **37**. Instead, the adaptive equalizer input signal is applied directly to the clocked adder **37**-**0** as a summand input signal, its weight W_{0 }being presumed to have unit value, +1. The filter coefficients computer **49** computes pre-echo compensation signals respectively applied as summand input signals to the clocked adders **37**-**1**, **37**-**2**, . . . **37**-(N−2), **37**-(N−1) and the final unclocked digital adder **37**-N. The

**49** in which these pre-echo compensation signals are generated. The filter coefficients computer **49** includes a PIPO register **490** for CIR, a multiplexer **491**, a bank **492** of ROM multipliers, a further PIPO register **493** for CIR, a bank **494** of adders, circuitry **495** for multiplying post-echo terms in PIPO register **493** by −1, a bank **497** of ROM dividers, and component computers **496** and **498**. These elements correspond in structure and operation to the PIPO register **410** for CIR, the multiplexer **411**, the bank **412** of ROM multipliers, the further PIPO register **413**, the bank **414** of adders, the circuitry **415** for multiplying post-echo terms in PIPO register **413** by −1, the bank **417** of ROM dividers, and the component computers **416** and **418** that **41**. The cursor term of the CIR stored in the further PIPO register **493** is applied as a W_{0 }weighting coefficient to the multiplier **36**-**0** in **22** or **23** of

A component computer **499** computes the discrete Fourier transform (DFT) of input signal samples stored in the buffer memory **35**. This DFT is supplied to a bank **49**A of read-only-memory digital multipliers connected for multiplying this DFT term-by-corresponding-term by the quotient DFT supplied from the bank **497** of ROM dividers. The product output signals of the bank **49**A of ROM multipliers generate the DFT of pre-echoes in the input signal. The DFTs that are multiplied together term-by-corresponding-term by the bank **49**A of ROM multipliers are zero-extended before that procedure, so that the product DFT can exhibit growth in the pre-echo terms without their wrapping around. A component computer **49**B computes the inverse discrete Fourier transform (I-DFT) of the product DFT to recover samples of the pre-echoes in the input signal. These samples are multiplied by minus one in circuitry **49**C to generate pre-echo compensation signals respectively applied as summand input signals to the clocked adders **37**-**1**, **37**-**2**, . . . **37**-(N−2), **37**-(N−1) and the final unclocked digital adder **37**-N. After the application of these pre-echo compensation signals, these summand signals are zero-valued until P symbol epochs go by.

The term-by-corresponding-term multiplication of the DFT of input signal by the quotient DFT in the **49**A of ROM multipliers in the filter coefficients computer **49** of the **36** of ROM multipliers in the **49** take advantage of the fact that partially de-echoed signal with most of the post-echoes canceled is available in the registers **37** and **50**.

**49**. The **49** differs from the FIG. **10** filter coefficients computer **49**. Instead of the CIR in the PIPO register **490** being forwarded every P symbol epochs to the further PIPO register **493** for CIR, the CIR from the PIPO register **490** is routed through a bank **49**D of ROM multipliers for partial normalization before application to a further PIPO register **49**E for CIR. At times other than loading the further PIPO register **49**E, the bank **49**D of ROM multipliers is operated in conjunction with the PIPO register **490**, the multiplexer **491** and the bank **494** of adders in the same way as the bank **492** of ROM multipliers is operated in conjunction with those elements in the **49**.

The pre-echo terms of the CIR from the PIPO register **490** are multiplied by 1/W_{0}, the reciprocal of the cursor term, in their transfer to the further PIPO register **49**E via the bank **49**D of ROM multipliers. The cursor term of the CIR from the PIPO register **490** is multiplied by unity in its transfer to the further PIPO register **49**E via the bank **49**D of ROM multipliers. (Alternatively, the cursor term of the CIR can be transferred directly from the PIPO register **490** to the further PIPO register **49**E without passage through a ROM multiplier in the bank **49**D of ROM multipliers, of course.) The cursor term of the CIR temporarily stored in the further PIPO register **49**E is applied as a W_{0 }weighting coefficient to the multiplier **36**-**0** in **22** or **23** of

The post-echo terms of the CIR from the PIPO register **490** are multiplied by unity in their transfer to the further PIPO register **49**E via the bank **49**D of ROM multipliers, or are transferred directly to the further PIPO register **49**E without passage through the bank **49**D of ROM multipliers. Alternatively, the post-echo terms of the CIR from the PIPO register **490** can instead be multiplied by 1/W_{0}, the reciprocal of the cursor term, in their transfer to the further PIPO register **49**E via the bank **49**D of ROM multipliers. **49**E are multiplied by minus one in circuitry **49**F. This generates the weighting coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(N−1)}, W_{+N }and W_{+(N+1) }that the digital multipliers **40**-**1**, **40**-**2**, **40**-**3**, . . . **40**-(N−1), **40**-N and **40**-(N+1) respectively use as their multiplier input signals.

In some variants of the **490** are multiplied by minus unity or by −1/W_{0 }in their transfer to the further PIPO register **49**E via the bank **49**D of ROM multipliers. The post-echo terms of the CIR from the further PIPO register **49**E are then applied directly to the digital multipliers **40**-**1**, **40**-**2**, **40**-**3**, . . . **40**-(N−1), **40**-N and **40**-(N+1) as the weighting coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(N−1)}, W_{+N }and W_{+(N+1)}. These variants of the **49**F for multiplying by minus unity the post-echo terms of the CIR from the further PIPO register **49**E.

A component computer **49**G computes the DFT of just the pre-echo portion of the CIR as it appears in normalized form in the further PIPO register **49**E. The terms of this DFT supply the multiplicand input signals for the bank **49**A of ROM multipliers in **49**H computes the discrete Fourier transform (DFT) of input signal samples stored in the final P stages of the chain-of-adders register **37** and in the SIPO register **50**. The terms of this DFT supply the multiplier input signals for the bank **49**A of ROM multipliers in **49**G. The product output signals of the bank **49**A of ROM multipliers generate the DFT of pre-echoes in the input signal. The DFTs that are multiplied together term-by-corresponding-term by the bank **49**A of ROM multipliers are zero-extended before that procedure, so that the product DFT can exhibit growth in the pre-echo terms without their wrapping around. The component computer **49**B computes the inverse discrete Fourier transform (I-DFT) of the product DFT to recover samples of the pre-echoes in the input signal. These samples are multiplied by minus one in circuitry **49**C to generate pre-echo compensation signals respectively applied as summand input signals to the clocked adders **37**-**1**, **37**-**2**, . . . **37**-(N−2), **37**-(N−1) and the final unclocked digital adder **37**-N. After the application of these pre-echo compensation signals, these summand signals are zero-valued until P symbol epochs go by.

In some variants of the **490** are multiplied by −1/W_{0 }in their transfer to the further PIPO register **49**E via the bank **49**D of ROM multipliers. In these variants the samples of the I-DFT from the component computer **49**B are pre-echo compensation signals respectively applied directly as summand input signals to the clocked adders **37**-**1**, **37**-**2**, . . . **37**-(N−2), **37**-(N−1) and the final unclocked digital adder **37**-N. These variants of the **49**C for multiplying these samples by minus unity.

Another variation of the equalization arrangement described with reference to **10** and **11** replaces the respective four SIPO registers **50** in the adaptive equalizers **16**, **17**, **20** and **21** with a shared SIPO register serially supplied input samples from the symbol synchronizer **24**. The equalized data symbols stored in this shared SIPO register are used to extend the samples available to the component computer **49**H from each of the four chain-of-adders registers **37**.

**37** for suppressing pre-echo components is extended with a preceding chain-of-adders register **51** for suppressing short-delay post-echo components as well. **51** comprising a plurality (M−1) in number of clocked digital adders **51**-**1**, **51**-**2**, **51**-**3**, . . . **51**-(M−2) and **51**-(M−1) connected, in reverse order of their numeric suffixes, in chain after a clocked data latch **51**-M. **52** of read-only-memory digital multipliers, M in number, each supplied signal delayed baseband DTV signal from the buffer memory **35** as a multiplicand input signal. A multiplier **52**-M in the bank **52** is connected to supply its product output to the clocked data latch **51**-M as input signal thereto. The clocked digital adders **51**-**1**, **51**-**2**, **51**-**3**, . . . **51**-(M−2) and **51**-(M−1) can be modified by respective summands other than the summands received via the chain connection. Multipliers **52**-**1**, **52**-**2**, **52**-**3**, . . . **52**-(M−2) and **52**-(M−1) in the bank **52** are connected to supply their respective product output signals to the clocked digital adders **51**-**1**, **51**-**2**, **51**-**3**, . . . **51**-(M−2) and **51**-(M−1) as their respective other summands. **43** connects to a filter coefficients computer **53** in **53**.

Delayed baseband DTV signal supplied from the buffer memory **35** in **54** of digital multipliers, N in number. These multipliers can be constructed in read-only memory, so as to exhibit minimal delay. Digital multipliers **54**-**0**, **54**-**1**, **54**-**2**, . . . **54**-(N−2), **54**-(N−1) and **54**-N in the bank **54** weight the adaptive equalizer input signal by respective weights W_{0}, W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }to generate products applied as summands to clocked digital adders **55**-**0**, **55**-**1**, **55**-**2**, . . . **55**-(N−2), **55**-(N−1) and **55**-N, respectively, which adders are N in number and are connected to form a chain-of-adders register **55**. The bank **54** of digital multipliers and the chain-of-adders register **55** shown in **36** of digital multipliers and the chain-of-adders register **37** in the **55**-**0** in this register **55** is connected to receive, as a further summand input thereto, the sum output signal from the final clocked digital adder **51**-**1** in the chain-of-adders register **51** depicted in **52** and **54** of digital multipliers combine with the cascaded chain-of-adders registers **51** and **55** to form an FIR filter. This FIR filter is used to suppress pre-echoes and short-delay post-echoes in the adaptive equalizer output signal supplied from the final clocked digital adder **55**-N in the chain-of-adders register **55** depicted in **55**-N in the chain-of-adders register **55** is connected to apply its output signal as serial input signal to a serial-in, parallel-out (SIPO) register **56**.

In **57** responds to each successive sample of the equalizer output signal supplied from the symbol synchronizer **24** to generate, a purality L symbol epochs later, a corresponding estimate of the symbol actually transmitted. A bank **58** of digital multipliers, N+1 in number, weights these estimates of the symbols actually transmitted, for application to taps in the chain-of-adders register **55**. The digital multipliers **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N−1), **58**-N and **59**-(N+1) in the bank **58** can be constructed in read-only-memory to minimize their latent delay. In **59** responds to each successive sample of the equalizer output signal from the symbol synchronizer **24** to supply a similar sample, L symbol epochs later, to the decision-feedback error detector **45**. Post-echoes with less than L symbol epochs delay are denominated “short-delay” post-echoes in this specification and its accompanying drawing. Post-echoes with at least L symbol epochs delay are denominated “long-delay” post-echoes.

The digital multipliers **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N−1), **58**-N and **58**-(N+1) weight each estimate by respective weights W_{+(L+1)}, W_{+(L+2)}, W_{+(L+3)}, . . . W_{+(L+N−1)}, W_{+(L+N) }and W_{+(L+N+1)}. These weights are computed by the filter coefficients computer **53** and are periodically updated. The filter coefficients computer **53** also computes and periodically updates the weights W_{0}, W_{1}, W_{2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }that **54**-**0**, **54**-**1**, **54**-**2**, . . . **54**-(N−2), **54**-(N−1) and **54**-N in the bank **54** use as their respective multiplier input signals. Furthermore, the computer **53** also computes and periodically updates the weighting coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(M−2)}, W_{+(M−1) }and W_{+M }that **52**-**1**, **52**-**2**, **52**-**3**, . . . **52**-(M−2), **52**-(M−1) and **52**-M use as their respective multiplier input signals.

**53**, which includes a PIPO register **530** for CIR, a multiplexer **531**, a bank **532** of ROM multipliers, a further PIPO register **533** for CIR, and a bank **534** of adders. These elements correspond in structure and operation to the PIPO register **410** for CIR, the multiplexer **411**, the bank **412** of ROM multipliers, the PIPO register **434** for normalized CIR and the bank **414** of adders that **41** to include.

The long-delay post-echo terms of each normalized CIR temporarily stored in the CIR register **533** are multiplied by minus one in circuitry **535**. This generates the weighting coefficients W_{+(L+1)}, W_{+(L+2)}, W_{+(L+3)}, . . . W_{+(L+N−1)}, W_{+(L+N) }and W_{+(L+N+1) }that the **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N−1), **58**-N and **58**-(N+1) respectively use as their multiplier input signals. The hR filtering performed by the chain-of-adders register **55** and the bank **58** of digital multipliers cancels all long-delay post-echoes, but pre-echoes and short-delay post-echoes remain to be suppressed by FIR filtering. The portion of the FIR filtering performed by the chain-of-adders register **55** and the bank **54** of multipliers in **51** and the bank **52** of multipliers in

Accordingly, as **53** includes a component computer **536** for computing the DFT of each CIR temporarily stored in the CIR register **533** to determine the channel impulse response in the frequency domain. This DFT provides the divisor terms for a bank **537** of ROM dividers that perform a term-by-corresponding-term division process in the frequency domain, which division process corresponds to a de-convolution process in the time domain. Any terms of the DFT of the normalized CIR that the component computer **536** computes that would be smaller than a prescribed value associated with acceptably small noise growth are augmented by the computer **536**, so as to replace the originally computed value with that prescribed value. I.e., the component computer **536** corresponds in structure and operation to the **416**.

The CIR register **533** is connected for supplying the only the pre-echo terms and short-delay post-echo terms of each normalized CIR temporarily stored therein to a component computer **538**. The computer **538** computes the DFT of the then-current pre-echo and short-delay post-echo terms for application to the bank **537** of ROM dividers as respective dividend terms for the term-by-corresponding-term division process in the frequency domain. The quotients from the bank **537** of ROM dividers specify a quotient DFT. A component computer **539** is connected to receive this quotient DFT from the bank **437** of ROM dividers and computes the inverse discrete Fourier transform (I-DFT) of the quotient DFT to synthesize the synthetic normalized CIR that is the impulse response in the time-domain that is to be compensated against by the FIR filtering. The echo terms of this synthetic normalized CIR are multiplied by minus one in circuitry **53**A. The samples that succeed the sample associated with the cursor provide the weighting coefficients W_{+1}, W_{+2}, W_{+3}, . . . W_{+(M−2)}, W_{+(M−1) }and W_{+M }that **52**-**1**, **52**-**2**, **52**-**3**, . . . **52**-(M−2), **52**-(M−1) and **52**-M use as their respective multiplier input signals. The samples that precede the sample associated with the cursor provide the weighting coefficients W_{−1}, W_{−2}, . . . W_{−(N−2)}, W_{−(N−1)}, and W_{−N }that **54**-**1**, **54**-**2**, . . . **54**-(N−2), **54**-(N−1) and **54**-N use as their respective multiplier input signals.

**52** and **54** of multipliers are dispensed with. Instead, the summand terms for FIR filtering are supplied to the chain-of-adders registers **51** and **55** from a filter coefficients computer **60** that replaces the filter coefficients computer **53**.

**60**. The filter coefficients computer **60** includes a PIPO register **600** for CIR, a multiplexer **601**, a bank **602** of ROM multipliers, a further PIPO register **603** for CIR, a bank **604** of adders, circuitry **605** for multiplying long-delay post-echo terms in PIPO register **603** by −1, the bank **607** of ROM dividers, and component computers **606** and **608**. These elements correspond in structure and operation to the PIPO register **530** for CIR, the multiplexer **531**, the bank **532** of ROM multipliers, the further PIPO register **533** for CIR, the bank **534** of adders, the circuitry **535** for multiplying post-echo terms in PIPO register **533** by −1, the bank **537** of ROM dividers, and the component computers **536** and **538** that **53**.

The cursor term of the CIR temporarily stored in the further PIPO register **603** is applied as a W_{0 }weighting coefficient to the multiplier **54**-**0** in **22** or **23** of **603** are multiplied by minus one in circuitry **605**. This generates the weighting coefficients W_{+(L+1)}, W_{+(L+2)}, W_{+(L+3)}, . . . W_{+(L+N−1)}, W_{+(L+N) }and W_{+(L+N−1) }that the **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N−1), **58**-N and **58**-(N+1) respectively use as their multiplier input signals.

A component computer **609** computes the discrete Fourier transform (DFT) of input signal samples stored in the buffer memory **35**. This DFT is supplied to a bank **60**A of read-only-memory digital multipliers connected for multiplying this DFT term-by-corresponding-term by the quotient DFT supplied from the bank **607** of ROM dividers. The product output signals of the bank **60**A of ROM multipliers generate the DFT of pre-echoes and short-delay post-echoes in the input signal. The DFTs that are multiplied together term-by-corresponding-term by the bank **60**A of ROM multipliers are zero-extended before that procedure, so that the product DFT can exhibit growth in the pre-echo and short-delay post-echo terms without their wrapping around. A component computer **60**B computes the inverse discrete Fourier transform (I-DFT) of the product DFT to recover samples of the pre-echoes and short-delay post-echoes in the input signal. These samples are multiplied by minus one in circuitry **60**C to generate pre-echo compensation signals and short-delay post-echo compensation signals. The pre-echo compensation signal samples preceding the cursor sample are applied as summand input signals to the clocked adders **55**-**1**, **55**-**2**, . . . **55**-(N−2), **55**-(N−1) and **55**-N, respectively. The short-delay post-echo compensation signal samples succeeding the cursor sample are applied as summand input signals to the clocked adders **51**-**1**, **51**-**2**, **51**-**3**, . . . **51**-(M−2) and **51**-(M−1) and to clocked data latch **51**-M for subsequent application to the clocked adder **51**-(M−1). In the intervening sample times until the contents of the SIPO register **56** are completely updated again, the chain-of-adders register **55** is conditioned not to accept any updating of its contents from external summand inputs, but is clocked as a shift register to reload the SIPO register **56**.

The term-by-corresponding-term multiplication of the DFT of input signal by the quotient DFT in the **60**A of ROM multipliers in the filter coefficients computer **60** of the adaptive equalizer shown in **52** of ROM multipliers in **54** of ROM multipliers in **60** take advantage of the fact that partially de-echoed signal with most of the post-echoes canceled is available in the registers **51** and **55**.

**60**. The **60** differs from the **60** in that instead of the CIR in the PIPO register **600** being forwarded every P symbol epochs to the further PIPO register **603** for CIR, the CIR from the PIPO register **600** is routed through a bank **60**D of ROM multipliers for partial normalization before application to a further PIPO register **60**E for CIR. At times other than loading the further PIPO register **60**E, the bank **60**D of ROM multipliers is operated in conjunction with the PIPO register **600**, the multiplexer **601** and the bank **604** of adders in the same way as the batik **602**, of ROM multipliers is operated in conjunction with those elements in the **60**.

The pre-echo terms and the short-delay post-echo terms of the CIR from the PIPO register **600** are multiplied by 1/W_{0}, the reciprocal of the cursor term, in their transfer to the further PIPO register **60**E via the bank **60**D of ROM multipliers. The cursor term of the CIR from the PIPO register **600** is multiplied by unity in its transfer to the further PIPO register **60**F via the bank **60**D of ROM multipliers. (Alternatively, the cursor term of the CIR can be transferred directly from the PIPO register **600** to the further PIPO register **60**E without passage through a ROM multiplier in the bank **60**D of ROM multipliers,) The cursor term of the CIR temporarily stored in the further PIPO register **60**E is applied as a W_{0 }weighting coefficient to the multiplier **54**-**0** in **22** or **23** of

The long-delay post-echo terms of the CIR from the PIPO register **600** are multiplied by unity in their transfer to the further PIPO register **60**E via the bank **60**D of ROM multipliers, or are transferred directly to the further PIPO register **60**E without passage through the bank **60**D of ROM multipliers. Alternatively, the long-delay post-echo terms of the CIR from the PIPO register **600** can instead be multiplied by 1/W_{0}, the reciprocal of the cursor term, in their transfer to the further PIPO register **60**E via the bank **60**D of ROM multipliers. **60**E are multiplied by minus one in circuitry **60**F. This generates the weighting coefficients W_{+(L+1)}, W_{+(L+2)}, W_{+(L+3)}, . . . W_{+(L+N−1)}, W_{+(L+N) }and W_{+(L+N+1) }that the **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N-−1), **58**-N and **58**-(N+1) respectively use as their multiplier input signals.

In some variants of the **600** are multiplied by minus unity or by −1/W_{0 }in their transfer to the further PIPO register **60**E via the bank **60**D of ROM multipliers. The long-delay post-echo terms of the CIR from the further PIPO register **60**E are then applied directly to the digital multipliers **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N−1), **58**-N and **59**-(N+1) as the weighting coefficients W_{+(L+1)}, W_{+(L+2)}, W_{+(L+3)}, . . . W_{+(L+N−1)}, W_{+(L+N) }and W_{+(L+N+1)}. These variants of the **60**F for multiplying by minus unity the long-delay post-echo terms of the CIR from the further PIPO register **60**E.

A component computer **60**G computes the DFT of just the pre-echo and short-delay post-echo portions of the CIR as they appear in normalized form in the further PIPO register **60**F. The terms of this DFT supply the multiplicand input signals for the bank **60**A of ROM multipliers in **60**H computes the discrete Fourier transform (DFT) of input signal samples stored in the final P stages of the chain-of-adders register **55** and in the SIPO register **56**. The terms of this DFT supply the multiplier input signals for the bank **60**A of ROM multipliers in **60**G. The product output signals of the bank **60**A of ROM multipliers generate the DFT of pre-echoes and short-delay post-echoes in the input signal. The DFTs that are multiplied together term-by-corresponding-term by the bank **60**A of ROM multipliers are zero-extended before that procedure, so that the product DFT can exhibit growth in the pre-echo terms and short-delay post-echo terms without wrap-around. The component computer **60**B computes the I-DFT of the product DFT to recover samples of the pre-echoes and short-delay post-echoes in the input signal. These samples are multiplied by minus one in circuitry **60**C to generate pre-echo compensation signals and short-delay post-echo compensation signals. The pre-echo compensation signal samples preceding the cursor sample are applied as summand input signals to the clocked adders **55**-**1**, **55**-**2**, . . . **55**-(N−2), **55**-(N−1) and **55**-N, respectively. The short-delay post-echo compensation signal samples succeeding the cursor sample are applied as summand input signals to the clocked adders **51**-**1**, **51**-**2**, **51**-**3**, . . . **51**-(M−2) and **51**-(M−1) and to clocked data latch **51**-M for subsequent application to the clocked adder **51**-(M−1). In the intervening sample times until the contents of the SIPO register **56** are completely updated again, the chain-of-adders register **55** is conditioned not to accept any updating of its contents from external summand inputs, but is clocked as a shift register to reload the SIPO register **56**.

In some variants of the **600** are multiplied by minus unity or by −1/W_{0 }in their transfer to the further PIPO register **60**E via the bank **60**D of ROM multipliers. The long-delay post-echo terms of the CIR from the further PIPO register **60**E are then applied directly to the digital multipliers **58**-**1**, **58**-**2**, **58**-**3**, . . . **58**-(N−1), **58**-N and **58**-(N+1) as the weighting coefficients W_{+(L+1)}, W_{+(L+2)}, W_{+(L+3)}, . . . W_{+(L+N−1)}, W_{+(L+N) }and W_{+(L+N+1)}. These variants of the **60**F for multiplying by minus unity the long-delay post-echo terms of the CIR from the further PIPO register **60**E.

**13** and **14** are squared by squarers **71** and **72**, respectively. The resulting squared samples are then differentially combined by a digital subtractor **73** to generate a digital difference signal with a strong 10.76 MHz component in its system function. The digital subtractor **73** is connected for supplying this digital difference signal to a digital-to-analog converter **74** for conversion to an analog signal with a strong 10.76 MHz component. The DAC **74** is connected for applying that analog signal as an input signal to a narrow bandpass filter **75** for selecting the 10.76 MHz component as first input signal to an automatic-frequency-control detector **76**. The narrow bandpass filter **75** typically includes a surface-acoustic-wave (SAW) filter designed to exhibit minimal change in phase response near the center of its passband. Alternative ways of supplying the narrow bandpass filter **75** all analog signal with a strong 10.76 MHz component are known, envelope detection of DTV IF signal being one of them.

The AFC detector **76** supplies automatic-frequency-control signal for adjusting the frequency of oscillations generated by a master oscillator **77** for clocking digital sampling throughout the DTV receiver. The master oscillator **77** is typically a crystal-controlled-oscillator having a natural oscillation frequency that is a multiple of 10.76 MHz. The average-axis or “zero” crossings of the sinusoidal oscillations of the master oscillator **77** are detected by a zero-crossing detector **78**, which typically comprises a symmetrical clipper followed by a differentiator and then a pulse rectifier. The pulses generated by the zero-crossing detector **78** when the sinusoidal oscillations of the master oscillator **77** cross average axis are supplied as input trigger signal to a frequency-divider chain **79** of flip-flops. The frequency-divider chain **79** divides the frequency of the master oscillator **77** oscillations to a nominally 10.76 MHz submultiple thereof for application to the AFC detector **76** as its second input signal. The AFC detector **76** compares the frequencies of its first and second input signals over time to generate the AFC error signal used to trim the frequency of the master oscillator **77** oscillations. This trimming is done to regulate the second input signal supplied to the AFC detector **76** so as to match the 10.76 MHz symbol rate of the DTV transmitter from which the **77** oscillations to a predetermined multiple of the 10.76 MHz symbol rate of the DTV signal currently being received. The frequency-divider chain **79** supplies signals related to the frequency of the master oscillator **77** oscillations for clocking various digital operations in the **79** to the ADC **12** for controlling digital sampling therein.

The pulses generated by the zero-crossing detector **78** when the sinusoidal oscillations of the master oscillator **77** cross average axis are supplied as count input to a zero-crossing counter **80**. The zero-crossing counter **80** counts the number of average-axis of the sinusoidal oscillations of the master oscillator **77** over one or more data frames, and the counter **80** count output or portions thereof can be decoded for controlling many operations in the DTV receiver. For example, the count range in which an echo-cancellation reference signal is expected to occur is decoded for controlling the training signal gate **42**. It is generally convenient to construct the counter **80** so that, in respective portions of its complete count, the counter **80** modularly counts the number of samples per symbol epoch, the number of symbol epochs per data segment, the number of data segments per data field, and the number of data fields per data frame. This simplifies decoding portions of the full count for respective operations in the DTV receiver.

**80** responsive to the occurrence of the PN511 sequence in the first data segment of each field of a DTV signal transmitted in accordance with the ATSC Digital Television Standard Document A/53, Annex D. A PN511 match filter **81** is connected for responding to the baseband DTV signal supplied from the symbol synchronizer **24** as its output signal. The response of the PN511 match filter **81** exhibits a very high peak when the PN511 sequence occurs in the first data segment of each data field. A threshold detector **82** responds to the occurrence of this very high peak to generate a reset signal, which is supplied to the counter **80**. Responsive to this reset signal the binary counter stages in the counter **80** are jam loaded to cause a prescribed count output from the counter **80**. More sophisticated circuits for periodically resetting the counter **80** to prescribed count output are possible, of course, as known or obvious to those skilled in the art.

Up to this point, this specification has chiefly concerned itself with the tracking operation of the adaptive equalizers **16**, **17**, **20** and **21**, after the initial values of the weighting coefficients in these digital filters have been established. The initialization of the weighting coefficients in these digital filters will depend on whether or not changes are made in the ATSC Digital Television Standard set forth in Document A/53 (Annex D). Modifications of Annex D to include repetitive-PN511 or repetitive-PN1023 sequences with baud rate symbols at either +5 or −5 modulation levels in each data field have been proposed. With such modifications the computation of initial values of the weighting coefficients for the adaptive equalizers **16**, **17**, **20** and **21** can be done by DFT methods adapted from those described by Charles Dietrich and Arthur Greenberg in U.S. Pat. No. 5,065,242 issued 23 Aug. 1994 and titled “DEGHOSTING APPARATUS USING PSEUDORANDOM SEQUENCES”. More specific description of the computation of initial values of the weighting coefficients for adaptive equalizers from repetitive-PN511 sequences can be found in U.S. patent application No. 20010033341 published Oct. 25, 2001; titled “GHOST CANCELLATION REFERENCE SIGNALS FOR BROADCAST DIGITAL TELEVISION SIGNAL RECEIVERS AND RECEIVERS FOR UTILIZING THEM”; and filed by A. L. R. Limberg on Jan. 18, 2001. Correspondingly titled U.S. Pat. No. 6,816,204 issued Nov. 9, 2004. More specific description of the computation of initial values of the weighting coefficients for adaptive equalizers from repetitive-PN1023 sequences can be found in U.S. patent application No. 20020051087 published May 2, 2002; titled “REPETITIVE-PN1023-SEQUENCE ECHO-CANCELLATION REFERENCE SIGNAL FOR SINGLE-CARRIER DIGITAL TELEVISION BROADCAST SYSTEMS”; and filed by A. L. R. Limberg, J. D. McDonald and C. B. Patel on Jul. 11, 2001. Correspondingly titled U.S. Pat. No. 6,768,517 issued Jul. 27, 2004. The initialization of the weighting coefficients of the adaptive equalizers **16**, **17**, **20** and **21** when receiving DTV signals transmitted in accordance with Document A/53, Annex D is a substantially more challenging task.

**43** in a less expensive embodiment of the **83** is connected for responding to a range of counter **80** output count corresponding to the interval when PN511 sequence is expected to occur, to generate a control signal. This control signal conditions the training signal gate **42** to extract those portions of the responses of the synchronous detectors **13** and **14** that occur at the times the PN511 sequence occurs in the initial segment of each data field. A computer **431** in the CIR extraction circuitry **43** zero-extends the extracted PN511 sequence to at least 1024 samples and thereafter computes the DFT of the-zero-extended sequence. A bank **432** of read-only memories divides this DFT, term by corresponding term by the DFT of a Nyquist-filtered ideal PN511 sequence to generate the initial CIR in the frequency domain. A computer **433** computes the inverse discrete Fourier transform of the initial CIR in the frequency domain to generate the initial CIR in the time domain from which the initial weighting coefficients for the adaptive equalizers **16**, **17**, **20** and **21** are determined.

This initial CIR is close enough that the customary operating procedures can establish close tracking within a few thousand symbols. However, in order to accomplish this, the trellis decoder **25** has to make very few errors or none at all, so that the decisions fed in to the decision feedback part of the decoder are cumulatively correct. This means that, when multipath reception conditions are adverse, establishing tracking (or re-establishing tracking, if it be lost) is apt to be slow or may not be possible. Start-up performance of the DTV receiver using the single PN511 sequence is improved by limiting the amount of new data processed in each successive set of DFT computations to only 256 symbols or less. The improvement is as compared to the processing of 2048 symbols or so of new data that is customary once close tracking has been established.

**84** is connected for generating a control signal in response to ranges of counter **80** output count that overlap the conclusions of data, fields and extend into the beginnings of succeeding data fields when PN511 sequences should occur. This control signal conditions a widened training signal gate **84** to selectively reproduce concurrent portions of the baseband DTV signals that the synchronous detector **13** and the synchronous detector **14** supply. The widened training signal gate **84** is connected for supplying these reproduced portions of the baseband DTV signal as input signal to a computer **86**. The computer **86** is used for computing the DFT of each sequence selectively reproduced by the gate **84** during an interval 2048 or 4096 symbol epochs long.

A further computer **87** performs an auto-correlation of this DFT to generate a CIR in the frequency domain. If this CIR in the frequency domain were converted to Euler form, a portion of its amplitude spectrum would be similar to that of the CIR generated from the bank **432** of ROMs as converted to Euler form. However, this portion of the amplitude spectrum of the CIR generated by the computer **87** is relatively free of the noise caused by edge effects that would reduce the SNR of the amplitude spectrum of the CIR generated from the bank **432** of ROMs in response to just the PN511 sequence. The phases of the corresponding spectral components would differ owing in the time differential in when the samples are taken for computation of the CIR generated by the computer **87** and the CIR generated from the bank **432** of ROMs. A computer **88** converts the CIR generated by the computer **87** in the frequency domain to recover the amplitude spectrum portion of that CIR expressed in Euler form. A computer **89** converts the CIR generated from the bank **432** of ROMs in the frequency domain to recover the phasing portions of that CIR expressed in Euler form. In accordance with an aspect of the invention as embodied in preferred forms of DTV receiver, a computer **8**A combines the amplitude spectrum portion of the CIR generated by the computer **87** with the corresponding phasing portion of the CIR generated from the bank **432** of ROMs. This is done so as to generate a CIR in the frequency domain that is expressed in Euler form and that has about a 3 dB better SNR than the CIR generated from the bank **432** of ROMs in the frequency domain would have if converted to Euler form. The computer **8**A subsequently converts to Cartesian form this initial CIR in the frequency domain, which CIR is expressed in Euler form and has better SNR. A computer **8**B computes the inverse discrete Fourier transform of the initial CIR in the frequency domain, as expressed in Cartesian form to generate the initial CIR in the time domain from which the initial weighting coefficients for the adaptive equalizers **16**, **17**, **20** and **21** are determined.

In a variant of the **84** is interposed between the synchronous detection circuitry and the training signal gate **42**. A variant of the

**16** and **17**, the shim delay circuitry **18** and **19**, and the adaptive synchronous equalizers **20** and **21**. Adaptive equalizers **92** and **93** perform a novel species of synchronous equalization in the **92** has the same number of digital multipliers as each of the adaptive synchronous equalizers **16** and **20**. The adaptive synchronous equalizer **93** has the same number of digital multipliers as each of the adaptive synchronous equalizers **17** and **21**. So, the

**13** remain time-division multiplexed for application as input signal to the adaptive synchronous equalizer **92**. The adaptive synchronous equalizers **16** and **20** in the **92** is the same as the respective sets of weighting coefficients in the **16** and **20**, but these weighting coefficients are applied to kernel taps two half-symbol-epochs apart, rather than a single-symbol-epoch apart. The intervening kernel taps in the adaptive synchronous equalizer **92** have zero-valued weighting coefficients associated with them. Selector circuitry **94** selects to shin delay circuitry **95** baud-rate alternate samples of adaptive synchronous equalizer **92** response to the samples at nτ times of the baseband DTV signal from the synchronous detector **13**. The shim delay circuitry **95** delays these samples one-half a symbol epoch for application to the I-and-Q-extraction circuitry **22** as a replacement for the response of the **20**. Selector circuitry **96** selects to the I-and-Q-extraction circuitry **23** baud-rate alternate samples of adaptive synchronous equalizer **92** response to the samples at (n+0.5)τ times of the baseband DTV signal from the synchronous detector **13**. In **96** replaces the response of the **16** as input signal for the I-and-Q-extraction circuitry **23**.

In **14** remain time-division multiplexed for application as input signal to the adaptive synchronous equalizer **93**. The adaptive synchronous equalizers **17** and **21** in the **93** is the same as the respective sets of weighting coefficients in the **17** and **210**, but these weighting coefficients are applied to kernel taps two half-symbol-epochs apart, rather than a single-symbol-epoch apart. The intervening kernel taps in the adaptive synchronous equalizer **93** have zero-valued weighting coefficients associated with them. Selector circuitry **97** selects to shim delay circuitry **98** baud-rate alternate samples of adaptive synchronous equalizer **93** response to the samples at nτ times of the baseband DTV signal from the synchronous detector **14**. The shim delay circuitry **98** delays these samples one-half a symbol epoch for application to the I-and-Q-extraction circuitry **22** as a replacement for the response of the **21**. Selector circuitry **99** selects to the I-and-Q-extraction circuitry **23** baud-rate alternate samples of adaptive synchronous equalizer **93** response to the samples at (n+0.5)τ times of the baseband DTV signal from the synchronous detector **14**. In **99** replaces the response of the **17** as input signal for the I-and-Q-extraction circuitry **23**.

The DTV receiver as shown in

**12** connected to digitize IF DTV signal from the tuner **10** to be applied as a respective input signal to each of two synchronous detectors **100** and **101**. The two synchronous detectors **100** and **101** are connected for performing respective synchronous detection procedures nominally in phase with the suppressed carrier of the IF DTV signal and in quadrature with the suppressed carrier of the IF DTV signal, respectively. In actual practice the two synchronous detectors **100** and **101** are usually constructed as parts of a shared structure including digital complex multiplier apparatus. The digitized IF DTV signal is supplied to a phase-splitter in this shared structure for conversion to a complex digital multiplicand for the digital complex multiplier apparatus. A digital controlled oscillator (DCO) **102** is connected for supplying the digital complex multiplier apparatus with a digital complex multiplier signal. This digital complex multiplier signal is composed of two digital carrier signals orthogonal in their respective phasings. Nominally, one of these two digital carrier signals is in phase with the suppressed carrier of the digitized IF DTV signal, and the other of these two digital carrier signals is advanced 90° respective to the suppressed carrier of the digitized IF DTV signal. The demodulation results for the synchronous detectors **100** and **101** are the real and imaginary components, respectively, of the complex product signal generated by the digital complex multiplier apparatus in their shared structure. The demodulation results for the synchronous detectors **100** and **101** are each composed of a set of samples of baseband DTV signals generated at twice Nyquist rate.

The samples of the imaginary component Q of the baseband DTV signal generated by the synchronous detector **101** are supplied to an AFPC loop filter **103**, which responds with an automatic-phase-and-frequency-control (AFPC) signal for the DCO **102**. The AFPC loop formed by feeding back the AFPC signal to the DCO **102** is designed to control the phases of the orthogonal digital carrier signals to keep synchrodyning of the digitized IF signal by the synchronous detector **100** in phase with the suppressed carrier of the digitized IF DTV signal and to keep synchrodyning of the digitized IF signal by the synchronous detector **101** in quadrature with the suppressed carrier of the digitized IF DTV signal.

The samples of the real component I of the complex baseband DTV signal generated by the synchronous detector **100** are applied as input signal to a pilot suppression filter **104**. The pilot suppression filter **104** suppresses in its system response the zero-frequency component of the samples of the real-only baseband DTV signal I, which zero-frequency component usually is primarily attributable to synchronous detection of the pilot. The pilot suppression filter **104** is connected to apply its pilot-free response as input signal to an adaptive synchronous equalizer **105**. The AFPC loop for the DCO **102** is not routed through channel-equalization and echo-suppression filtering, so there is no need for such filtering to preserve the zero-frequency component attributable to synchronous detection of the pilot. The data slicing procedures in the quantizers of the adaptive synchronous equalizer **105** are simplified if the data slicing bills are symmetrically disposed above and below zero-signal level. The suppression of the zero-frequency pilot component of the response of an adaptive synchronous equalizer can be done just before quantization, but doing it before the application of input signal to the equalizer has a number of practical advantages known to equalizer designers. For example, the suppression of the zero-frequency pilot component of the input signal to the adaptive synchronous equalizer **105** reduces the dynamic range required for that signal, since fluctuation of that the zero-frequency pilot component no longer is a consideration in determining that dynamic range.

**105** having adjustable weighting coefficients applied to kernel taps two half-symbol-epochs apart, rather than a single-symbol-epoch apart. The intervening kernel taps in the adaptive synchronous equalizer **105** have zero-valued weighting coefficients associated with them. This procedure halves the number of multipliers used in the adaptive equalizer compared to spacing kernel taps a single-symbol-epoch apart in two adaptive synchronous equalizers, one for processing even samples at nτ times and the other for processing odd samples at (n+0.5τ) times.

Selector circuitry **106** selects to shim delay circuitry **107** baud-rate samples of adaptive synchronous equalizer **105** response to the alternate samples at nτ times of the baseband DTV signal from the synchronous detector **100**. The shim delay circuitry **107** delays these samples one-half a symbol epoch for application to the symbol synchronizer **24**. Selector circuitry **108** selects to the symbol synchronizer **24** baud-rate samples of adaptive synchronous equalizer **105** response to alternate samples at (n+0.5)τ times of the baseband DTV signal from the synchronous detector **100**. The symbol synchronizer **24** combines the equalized real (I) components of the baseband DTV signal supplied from the I shim delay circuitry **107** and from the selector circuitry **108** to generate the samples of the equalized real (I) component of baseband DTV signal applied to the trellis decoder **25**. The symbol synchronizer **24** combines the equalized real components such that the samples applied to the trellis decoder **25** are timed to occur at τ intervals that minimize inter-symbol interference (ISI). The circuitry in the **25** is similar to that in the

**15** AFPC loop. An adder **124** additively combines the responses of the synchronous detectors **13** and **14** to obtain a sum signal that is, in effect, a synchronous detection response for 90° digital carrier when reception is free from multipath distortion. This sum signal is supplied to the AFPC loop filter **27** for being lowpass filtered to generate AFPC signal for controlling the DCO **15**. In the **15** is not routed through the adaptive synchronous equalizers **16**, **17**, **20** and **21**. So, there is no need for the equalizers **16**, **17**, **20** and **21** to preserve the zero-frequency component attributable to synchronous detection of the pilot.

A pilot suppression filter **132** is connected for receiving as its input signal the synchronous detector **13** response. The pilot suppression filter **132** is connected for applying its response, which is essentially free of zero-frequency pilot component, as input signal to the adaptive synchronous equalizer **16** and as input signal to the shim delay circuitry **18**. A pilot suppression filter **133** is connected for receiving as its input signal the synchronous detector **14** response. The pilot suppression filter **133** is connected for applying its response, which is essentially free of zero-frequency pilot component, as input signal to the adaptive synchronous equalizer **17** and as input signal to the shim delay circuitry **19**.

In the **122** replaces the I-and-Q-extraction circuitry **22** and I-extraction circuitry **123** replaces the I-and-Q-extraction circuitry **23**. The I-extraction circuitry **122** can be constructed similarly to the I-and-Q-extraction circuitry **22** shown in **227** and the ROM multipliers **226**, **228** and **229**. The I-extraction circuitry **123** can constructed similarly to the I-and-Q-extraction circuitry **23** shown in **237** and the ROM multipliers **236**, **238** and **239**.

**15** AFPC loop. The further modifications in

**100** and **101** demodulating at 0° and 90° for the synchronous detectors **13** and **14** demodulating at 45° and 135°. As in the **102** is connected for supplying two digital carrier signals for implementing synchronous detection by the synchronous detectors **100** and **101**. Nominally, one of these two digital carrier signals is in phase with the suppressed carrier of the digitized IF DTV signal, and the other of these two digital carrier signals is advanced 90° respective to the suppressed carrier of the digitized IF DTV signal. The imaginary component Q of the baseband DTV signal generated by the synchronous detector **101** is supplied to an AFPC loop filter **103**, which responds with an AFPC signal for the DCO **102**. As in the **104** is connected for receiving, as its input signal, the I signal demodulated by the synchronous detector **100**. Another pilot suppression filter **109** of construction similar to the filter **104** is connected for receiving, as its input signal, the Q signal demodulated by the synchronous detector **101**. The samples of the baseband DTV signals from the pilot suppression filters **104** and **109** occurring at (n+0.5)τ times are applied as input signals to the adaptive equalizers **16** and **17**, respectively, in a 2:1 decimation procedure. The samples of the baseband DTV signals from the synchronous pilot suppression filters **104** and **109** occurring at nτ times are delayed 0.5τ by re-clocking in the shim delay circuitry **18** and **19**, respectively. The re-clocked samples from the shim delay circuitry **18** and from the shim delay circuitry **19** are applied as input signals to adaptive equalizers **20** and **21**, respectively, in a 2:1 decimation procedure. The re-clocking in the shim delay circuitry **18** and **19** temporally aligns the subsets of samples occurring at nτ times that are applied as input signals to the adaptive equalizers **20** and **21** with the subsets of samples occurring at (n+0.5)τ times that are applied as input signals to the adaptive equalizers **16** and **17**.

A surprising result in the operation of the DTV receiver modified as shown in **17** and **21** are automatically adapted so they perform as a Hilbert filter to supply samples at nτ times and at (n+0.5)τ times of de-echoed signal similar to those supplied by the adaptive equalizers **16** and **20**. This result confirms the capability of the **13** and **14** being rotated from the nominal demodulation axes.

A variant of the **101**. In this variant the adaptive equalizer **17** and the shim delay **19** are connected to receive the Hilbert filter response as their respective input signals, rather than to receive their input signals directly from the synchronous detector **101**. In this variant, the synchronous detector **100** response is delayed to compensate for the latent delay in the Hilbert filter, and the adaptive equalizer **16** and the shim delay **18** are connected to receive this delayed synchronous detector **100** response as their respective input signals, rather than to receive their input signals directly from the synchronous detector **100**.

**15**, which loop includes the adaptive synchronous equalization filtering. **15**, which loop excludes the adaptive synchronous equalization filtering. The bandwidth of the AFPC loop filter **27** can be chosen to be around 2 kHz. **102**, which loop excludes the adaptive synchronous equalization filtering. The bandwidth of the AFPC loop filter **103** can also be chosen to be around 2 kHz. A 2 kHz bandwidth for the AFPC loop of carrier recovery circuitry is common practice since it allows phase noise in the DTV receiver local oscillators to be tracked, but keeps the effects of data on the AFPC signal negligible. An AFPC loop of such narrow bandwidth restricts the range of frequency over which local oscillations can be pulled into phase-lock with the carrier of the IF DTV signal more than is desirable in a DTV receiver intended for the home market. For this reason, it is preferable that a Costas type of AFPC loop is used to develop AFPC signal for the DCO **15** in modifications of the DTV receivers of **20**, **22** and **23**. For the same reason, it is preferable that a Costas type of AFPC loop is used to develop AFPC signal for the DCO **103** in modifications of the DTV receivers in

**103**. The AFPC loop is preferably constructed using digital circuitry, but resembles an AFPC loop constructed using analog circuitry that is described in Subsection 10.2.3.2 “Channel filtering and VSB carrier recovery” of ATSC Document A/54 *Guide to the Use of the ATSC Digital Television Standard *published 4 Oct. 1995. In these modifications the AFPC loop filter **103** is replaced by an AFPC loop filter **116** with an integrator. The synchronous detector **101** is connected for applying the samples of the imaginary component Q of the baseband DTV signal that it generates to a multiplier **117** as a multiplicand input signal. The multiplier **117** is connected to supply its product output signal to the AFPC loop filter **116**. The synchronous detector **100** is connected for applying the samples of the real component I of the baseband DTV signal that it generates to a lowpass AFC loop filter **118** as input signal thereto. The AFC loop filter **118** is designed to exhibit lag in its response to the frequencies contained in the real component I of the baseband DTV signal which lag increases with frequency, from zero lag at zero-frequency to 90° at a frequency well above the filter **118** cut-off frequency of about 100 kHz. A polarity detector **119** is connected to receive the lowpass AFC loop filter **118** response as input signal. The polarity detector **119** is connected to supply its output signal to the multiplier **117** as a multiplier input signal.

Since the AFC loop filler **118** has a lowpass cut-off frequency of about 100 kHz, the spectral energy in demodulated data from the synchronous detector **100** and from any strong Johnson noise accompanying that demodulated data is strongly attenuated, presuming that those spectral components have reasonably uniform spectral distribution over the synchronous detector **100** baseband response. Most of the spectral energy in the synchronous detector **100** baseband response from any demodulation artifacts of an interfering co-channel NTSC analog television signal is also strongly attenuated. The demodulation artifacts from frequency components of the interfering co-channel NTSC analog television signal near its video carrier and in its upper sideband are especially attenuated. When the IF DTV signal carrier and oscillations from the DCO **102** are not in a phase-lock condition, the lowpass filter response from the AFC loop filter **118** will be predominately an I-channel beat signal. This I-channel beat signal is composed of the difference frequencies between the frequency of oscillations from the DCO **102** and frequencies of the IF DTV signal. The I-channel beat signal will be primarily at the difference frequency between the frequency of oscillations from the DCO **102** and the frequencies of the pilot component of the IF DTV signal. The polarity detector **119** generates a +1 multiplier input signal for the multiplier **117** at instants when the I-channel beat signal is zero-valued or positive. At instants when that I-channel beat signal is negative, the polarity detector **119** generates a +1 multiplier input signal for the multiplier **117**. I.e., when the IF DTV signal carrier and oscillations from the DCO **102** are not in a phase-lock condition, so the I-channel beat signal is generated, the multiplier input signal for the multiplier **117** is a square wave with a fundamental frequency equal to the I-channel beat frequency signal and with a phase that is retarded in proportion to that fundamental frequency. The polarity detector **119** provides an amplitude-limiter type of operation which is usually “captured” by the I-channel beat signal being the strongest component of the lowpass filter response from the AFC loop filter **118**.

When the IF DTV signal carrier and oscillations from the DCO **102** are not in a phase-lock condition, the synchronous detector **101** baseband response includes a Q-channel beat signal at the difference frequencies between the frequencies of the IF DTV signal pilot and oscillations from the DCO **102**. The multiplier **117** operates as a mixer for the I-channel beat signal and the Q-channel beat signal generating a product signal that includes a zero-frequency component. The polarity of this zero-frequency component indicates whether the frequency of the pilot in the final IF DTV signal is above or below the frequency of oscillations from the DCO **102**. The amplitude of this zero-frequency component is proportional to the frequency of the beat signal between the DCO **102** oscillations and the pilot in the final IF DTV signal. This zero-frequency component ramps the AFPC signal stored in the integrator in the AFPC loop filter **116** up or down in value, depending on the polarity of the zero-frequency component. This change in the AFPC signal adjusts the DCO **102** so the frequency of its oscillations draw closer to the frequency of the pilot in the IF DTV signal. Pull-in range is ±100 kHz, as set by the cut-off frequency of the lowpass AFC loop filter **118**.

When the IF DTV signal carrier and oscillations from the DCO **102** come into the phase-lock condition, the demodulated pilot in the real component I of the complex baseband DTV signal supplied by the synchronous detector **100** becomes a zero-frequency component of constant polarity. This zero-frequency component of constant polarity is the principal component of the lowpass filter response from the AFC loop filter **118**, and the polarity detector **119** responds to supply a constant-value multiplier input signal to the multiplier **117**. Multiplicative mixing operation by the multiplier **117** is accordingly discontinued, and the multiplier **117** reproduces in its product output signal the quadrature component Q of the complex baseband DTV signal supplied by the synchronous detector **101**. This product signal is supplied as the input signal to the AFPC loop filter **116** to complete a simple AFPC loop for controlling the DCO **102**. This loop adjusts the phase of oscillations from the DCO **102** to be in prescribed relationship with the phase of the pilot signal in the IF DTV signal that is to be synchronously detected.

The Costas loop can lock to pilot carrier in either of two phases. In one lock-up condition the demodulated DTV signal from the synchronous detector **100** has a zero-frequency component of positive polarity owing to the synchronous detection of pilot. In the other lock-up condition the demodulated DTV signal from the synchronous detector **100** has a zero-frequency component of negative polarity owing to the synchronous detection of pilot. Although in actual practice the direct component of the demodulated DTV signal from the synchronous detector **100** is customarily removed before supplying it to the adaptive synchronous equalizer **105**, it is desirable that the demodulated DTV signal invariably have a prescribed direction of modulation.

In the **100** is connected to supply the real component I of the complex baseband DTV signal to a multiplier **120** as a multiplicand input signal. The multiplier **120** is connected to receive the output signal of the polarity detector **119** as multiplier input signal. When the zero-frequency component in the demodulated DTV signal from the synchronous detector **100** has positive polarity, the polarity detector **119** supplies a +1 multiplier input signal to the multiplier **120**. This conditions the multiplier **120** to reproduce in its product output signal the demodulated DTV signal from the synchronous detector **100**. When the zero-frequency component in the demodulated DTV signal from the synchronous detector **100** has negative polarity, the polarity detector **119** supplies a −1 multiplier input signal to the multiplier **120**. This conditions the multiplier **120** to produce as its product output signal an inverted response to the demodulated DTV signal from the synchronous detector **100**. The product output signal supplied from the multiplier **120** has a prescribed sense of modulation and its zero-frequency pilot has a prescribed polarity. The multiplier **120** is connected to apply its product output signal as the input signal to the pilot suppression filter **104**.

In the **101** is connected to supply the quadrature component Q of the complex baseband DTV signal to a multiplier **121** as a multiplicand input signal. The multiplier **121** is connected to receive the output signal of the polarity detector **119** as multiplier input signal. When the zero-frequency component in the demodulated DTV signal from the synchronous detector **100** has positive polarity, the polarity detector **119** supplies a +1 multiplier input signal to the multiplier **121**. This conditions the multiplier **121** to supply a product output signal that reproduces the demodulated DTV signal from the synchronous detector **101**. When the zero-frequency component in the demodulated DTV signal from the synchronous detector **100** has negative polarity, the polarity detector **119** supplies a −1 multiplier input signal to the multiplier **121**. This conditions the multiplier **121** to invert the polarity of demodulated DTV signal from the synchronous detector **101** in its product output signal. The multiplier **121** is connected to apply its product output signal as input signal to the pilot suppression filter **109**.

**15**. In this modification a subtractor **125** is connected for receiving the output signal of the synchronous detector **13** as minuend input signal and for receiving the output signal of the synchronous detector **14** as subtrahend input signal. Responsive to these input signals the subtractor **125** develops a real component I signal with accompanying echoes as its difference output signal. An AFPC loop filter **126** with an integrator, a multiplier **127**, a lowpass AFC loop filter **128** and a polarity detector **129** in the **116** with an integrator, the multiplier **117**, the lowpass AFC loop filter **118** and the polarity detector **119** in the

In the **24** DTV receiver the synchronous detector **13** is connected to supply demodulated DTV signal to a multiplier **130** as a multiplicand input signal. The multiplier **130** is connected to receive the output signal of the polarity detector **129** as multiplier input signal. The multiplier **130** supplies, as its product output signal, a baseband DTV signal that invariably has a prescribed direction of modulation no matter whether originally being demodulated at 45° or at 225°. The multiplier **130** product output signal is applied as input signal to the pilot suppression filter **132**.

In the **24** DTV receiver the synchronous detector **14** is connected to supply DTV signal to a multiplier **131** as a multiplicand input signal. The multiplier **131** is connected to receive the output signal of the polarity detector **129** as multiplier input signal. The multiplier **131** supplies, as its product output signal, a baseband DTV signal that invariably has a prescribed direction of modulation no matter whether originally being demodulated at 135° or at 315°. The multiplier **131** product output signal is applied as input signal to the pilot suppression filter **133**.

The adaptive equalizers shown in

While the computers **86**, **416**, **418**, **496**, **498**, **499**, **49**G. **49**H, **606**, **608**, **609**, **60**G and **60**H for computing DFTs could be microcomputers, those skilled in the art of designing such computers will understand that other constructions are preferable. In part this is because frequency-domain data are never used except in term-by-term multiplication procedures, so there is no call for the onerous “bit-reverse step” in the computations involving DFTs. Furthermore, it is easy to implement multiple “butterflies” in hardware, with hard-wired sines and cosines for 32 butterflies, by way of example. Thus, a 4096-point DFT takes 2048/32=64 steps for the first pass, 32 steps for the second pass, 16 steps for the third pass, 8 steps for the fourth pass, 4 steps for the fifth pass, 2 steps for the sixth pass, and a single step for each of the seventh through twelfth passes, for a total of 132 total time steps. Supposing the clock is four times 10.76 MHz—i.e., 43 Mhz— in an inexpensive silicon implementation, the 132 steps for performing the 4096-point DFT take 3.1 microseconds. However, since the last 5 stages use less than half the butterflies, two pipelined DFT operations can be overlapped. Thus on average, each 4096-point transform takes, say, 2.5 microseconds. Sixteen DFTs in total are needed per block taking 2.5×16=40 microseconds to perform. There are also other steps including the divides, so the time for a complete calculation is approximately 60 microseconds. 2048 symbol epochs extend over about 200 microseconds. So, about 30% of the time for processing a 2048-symbol block is needed for the DFTs, leaving 70% of the time for the remainder of the calculations of equalizer coefficient updates. Pipelined DET is commercially available, though DTV receiver designs will usually use custom integrated circuitry. Similarly, while the computers **419**, **49**B, **60**B, **433** and **8**B for computing I-DFTs could be microcomputers, those skilled in the art of designing such computers will understand that other constructions are preferable.

More extensive experience with the properties of these adaptive equalizers gained from computer simulations shows that how the time-domain responses of the FIR and IIR portions of the adaptive channel-equalization and echo-suppression filtering join affects how well some ensembles of echoes can be suppressed. The time-domain responses of the FIR and IIR portions of the adaptive filtering were originally conceived of as abutting each other at or shortly past the cursor component associated with the principal signal. Overlapping the time-domain responses of the FIR and IIR portions of the adaptive filtering can improve echo suppression, however, so simply abutting brick-wall time-domain responses of the FIR and IIR portions of the adaptive channel-equalization and echo-suppression filtering is not the optimal way to join these time-domain responses. These time-domain responses are better joined if the separation of the normalized CIR into portions for computing the IIR weighting coefficients for feedback filtering and for computing the FIR weighting coefficients for feed-forward filtering is done using digital filtering that exhibits a gradual cross-over. The cross-over stretches from a shorter-delayed post-echo term of the normalized CIR that can be suppressed by the IIR filtering to a longer-delayed post-echo term of the normalized CIR. Preferably, from the standpoint of reducing multiplications in the adaptive channel-equalization and echo-suppression filtering, the shorter-delayed post-echo term is the shortest-delayed post-echo term of the normalized CIR that can be suppressed by the IIR filtering. The cross-over can extend over as many as several tens of symbol epochs.

In the early development of the adaptation algorithms, the post-echo portion of the synthetic normalized CIR that was beyond the extent of the FIR filtering kernel was ignored in the generation of weighting coefficients, even though components of that post-echo portion might have significant energy, especially if there were not a gradual cross-over when separating the normalized CIR into portions for computing IIR weighting coefficients and for computing FIR weighting coefficients. Adaptation algorithms that are currently preferred do not ignore the post-echo terms of the synthetic normalized CIR that have significant energy, but are beyond the extent of the FIR filtering kernel. Instead, these post-echo terms are complemented and used to augment the weighting coefficients of the IIR-filtering. This completes the suppression of significant-energy post-echo terms that are generated by the FIR filtering suppressing preceding terms with delayed, scaled responses to equalizer input signal that is not echo-free. This procedure reduces the problems of equalization being satisfactory when post-echo terms of the synthetic normalized CIR have significant energy, but are beyond the extent of the FIR filtering kernel. Accordingly, the cross-over between the portions of the normalized CIR, respectively used for computing FIR weighting coefficients and for computing IIR weighting coefficients, does not need to extend over as many tens of symbol epochs.

The adaptation procedures described with reference to

In the claims which follow, the definite article “the” is used for grammatical purposes other than for indicating antecedence. Where antecedence is intended in a claim, it is indicated by the adjective “said”.

Citas de patentes

Patente citada | Fecha de presentación | Fecha de publicación | Solicitante | Título |
---|---|---|---|---|

US3614673 | 28 May 1970 | 19 Oct 1971 | Bunker Ramo | Technique for utilizing a single pulse to set the gains of a transversal filter |

US4027258 | 1 Jun 1976 | 31 May 1977 | Xerox Corporation | Time domain automatic equalizer with frequency domain control |

US5065242 | 29 Jun 1990 | 12 Nov 1991 | General Electric Company | Deghosting apparatus using pseudorandom sequences |

US5251033 | 16 Mar 1992 | 5 Oct 1993 | Rca Thomson Licensing Corporation | D.C. responsive equalization for television transmission channel irregularities |

US5479449 | 4 May 1994 | 26 Dic 1995 | Samsung Electronics Co. Ltd. | Digital VSB detector with bandpass phase tracker, as for inclusion in an HDTV receiver. |

US5799037 | 27 Sep 1996 | 25 Ago 1998 | David Sarnoff Research Center Inc. | Receiver capable of demodulating multiple digital modulation formats |

US6377312 | 13 Ago 1999 | 23 Abr 2002 | Samsung Electronics Co., Ltd. | Adaptive fractionally spaced equalizer for received radio transmissions with digital content, such as DTV signals |

Citada por

Patente citante | Fecha de presentación | Fecha de publicación | Solicitante | Título |
---|---|---|---|---|

US7860200 | 12 Oct 2007 | 28 Dic 2010 | Harris Corporation | Communications system using adaptive filter that is selected based on output power |

US7864835 | 12 Oct 2007 | 4 Ene 2011 | Harris Corporation | Communications system using adaptive filter and variable delay before adaptive filter taps |

US7912118 | 21 Sep 2006 | 22 Mar 2011 | Her Majesty The Queen In Right Of Canada, As Represented By The Minister Of Industry, Through The Communications Research Centre Canada | Hybrid domain block equalizer |

US7948703 * | 20 Ene 2009 | 24 May 2011 | Marvell International Ltd. | Adaptive target optimization methods and systems for noise whitening based viterbi detectors |

US7957490 | 20 Sep 2007 | 7 Jun 2011 | Limberg Allen Leroy | Insertion of repetitive PN sequences into DTV data fields |

US8073046 | 14 Jun 2007 | 6 Dic 2011 | Zoran Corporation | Fast training equalization of a signal by using adaptive-iterative algorithm with main path phase correction |

US8081722 | 4 Abr 2008 | 20 Dic 2011 | Harris Corporation | Communications system and device using simultaneous wideband and in-band narrowband operation and related method |

US8094763 | 12 Oct 2007 | 10 Ene 2012 | Harris Corporation | Communications system using adaptive filter with adaptive update gain |

US8098781 | 12 Oct 2007 | 17 Ene 2012 | Harris Corporation | Communications system using adaptive filter with normalization circuit |

US8107572 | 12 Oct 2007 | 31 Ene 2012 | Harris Corporation | Communications system using adaptive filter for interference reduction |

US8121236 | 12 Oct 2007 | 21 Feb 2012 | Harris Corporation | Communications system using adaptive filter circuit using parallel adaptive filters |

US8165192 | 19 Oct 2007 | 24 Abr 2012 | Panasonic Corporation | Waveform equalizer |

US8204164 | 12 Oct 2007 | 19 Jun 2012 | Harris Corporation | Communications system using adaptive filter and selected adaptive filter taps |

US8291304 * | 18 Abr 2011 | 16 Oct 2012 | Kabushiki Kaisha Toshiba | Signal processing device, signal processing method, and signal reproducing apparatus |

US8451376 * | 24 Abr 2012 | 28 May 2013 | Silicon Laboratories Inc. | Automatic gain control (AGC) for analog TV signals using feed-forward signal path delay |

US20070064824 * | 21 Sep 2006 | 22 Mar 2007 | Xianbin Wang | Hybrid domain block equalizer |

US20110264983 * | 27 Oct 2011 | Kabushiki Kaisha Toshiba | Signal processing device, signal processing method, and signal reproducing apparatus | |

CN101958857B | 17 Jul 2009 | 24 Abr 2013 | 瑞昱半导体股份有限公司 | Communication signal receptor and signal processing method thereof |

Clasificaciones

Clasificación de EE.UU. | 375/232, 348/E05.084, 375/219, 375/350, 348/E05.108, 375/267 |

Clasificación internacional | H04L25/02, H04N5/44, H03H7/30, H04B1/10, H04L25/03, H04B7/02, H04B1/38, H04N5/21 |

Clasificación cooperativa | H04L25/0212, H04N5/211, H04L2025/03382, H04L2025/03445, H04N5/4401 |

Clasificación europea | H04N5/21A, H04L25/02C3 |

Eventos legales

Fecha | Código | Evento | Descripción |
---|---|---|---|

6 Nov 2009 | FPAY | Fee payment | Year of fee payment: 4 |

3 Ene 2014 | REMI | Maintenance fee reminder mailed | |

23 May 2014 | LAPS | Lapse for failure to pay maintenance fees | |

15 Jul 2014 | FP | Expired due to failure to pay maintenance fee | Effective date: 20140523 |

Girar